mirror_qemu/migration/dirtyrate.c

924 lines
26 KiB
C
Raw Normal View History

/*
* Dirtyrate implement code
*
* Copyright (c) 2020 HUAWEI TECHNOLOGIES CO.,LTD.
*
* Authors:
* Chuan Zheng <zhengchuan@huawei.com>
*
* This work is licensed under the terms of the GNU GPL, version 2 or later.
* See the COPYING file in the top-level directory.
*/
#include "qemu/osdep.h"
#include "qemu/error-report.h"
#include <zlib.h>
#include "hw/core/cpu.h"
#include "qapi/error.h"
#include "exec/ramblock.h"
#include "exec/target_page.h"
#include "qemu/rcu_queue.h"
#include "qemu/main-loop.h"
#include "qapi/qapi-commands-migration.h"
#include "ram.h"
#include "trace.h"
#include "dirtyrate.h"
#include "monitor/hmp.h"
#include "monitor/monitor.h"
#include "qapi/qmp/qdict.h"
#include "sysemu/kvm.h"
#include "sysemu/runstate.h"
#include "exec/memory.h"
#include "qemu/xxhash.h"
/*
* total_dirty_pages is procted by BQL and is used
* to stat dirty pages during the period of two
* memory_global_dirty_log_sync
*/
uint64_t total_dirty_pages;
typedef struct DirtyPageRecord {
uint64_t start_pages;
uint64_t end_pages;
} DirtyPageRecord;
static int CalculatingState = DIRTY_RATE_STATUS_UNSTARTED;
static struct DirtyRateStat DirtyStat;
static DirtyRateMeasureMode dirtyrate_mode =
DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING;
static int64_t dirty_stat_wait(int64_t msec, int64_t initial_time)
{
int64_t current_time;
current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
if ((current_time - initial_time) >= msec) {
msec = current_time - initial_time;
} else {
g_usleep((msec + initial_time - current_time) * 1000);
/* g_usleep may overshoot */
msec = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - initial_time;
}
return msec;
}
static inline void record_dirtypages(DirtyPageRecord *dirty_pages,
CPUState *cpu, bool start)
{
if (start) {
dirty_pages[cpu->cpu_index].start_pages = cpu->dirty_pages;
} else {
dirty_pages[cpu->cpu_index].end_pages = cpu->dirty_pages;
}
}
static int64_t do_calculate_dirtyrate(DirtyPageRecord dirty_pages,
int64_t calc_time_ms)
{
uint64_t increased_dirty_pages =
dirty_pages.end_pages - dirty_pages.start_pages;
/*
* multiply by 1000ms/s _before_ converting down to megabytes
* to avoid losing precision
*/
return qemu_target_pages_to_MiB(increased_dirty_pages * 1000) /
calc_time_ms;
}
void global_dirty_log_change(unsigned int flag, bool start)
{
system/cpus: rename qemu_mutex_lock_iothread() to bql_lock() The Big QEMU Lock (BQL) has many names and they are confusing. The actual QemuMutex variable is called qemu_global_mutex but it's commonly referred to as the BQL in discussions and some code comments. The locking APIs, however, are called qemu_mutex_lock_iothread() and qemu_mutex_unlock_iothread(). The "iothread" name is historic and comes from when the main thread was split into into KVM vcpu threads and the "iothread" (now called the main loop thread). I have contributed to the confusion myself by introducing a separate --object iothread, a separate concept unrelated to the BQL. The "iothread" name is no longer appropriate for the BQL. Rename the locking APIs to: - void bql_lock(void) - void bql_unlock(void) - bool bql_locked(void) There are more APIs with "iothread" in their names. Subsequent patches will rename them. There are also comments and documentation that will be updated in later patches. Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> Reviewed-by: Paul Durrant <paul@xen.org> Acked-by: Fabiano Rosas <farosas@suse.de> Acked-by: David Woodhouse <dwmw@amazon.co.uk> Reviewed-by: Cédric Le Goater <clg@kaod.org> Acked-by: Peter Xu <peterx@redhat.com> Acked-by: Eric Farman <farman@linux.ibm.com> Reviewed-by: Harsh Prateek Bora <harshpb@linux.ibm.com> Acked-by: Hyman Huang <yong.huang@smartx.com> Reviewed-by: Akihiko Odaki <akihiko.odaki@daynix.com> Message-id: 20240102153529.486531-2-stefanha@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2024-01-02 18:35:25 +03:00
bql_lock();
if (start) {
memory_global_dirty_log_start(flag);
} else {
memory_global_dirty_log_stop(flag);
}
system/cpus: rename qemu_mutex_lock_iothread() to bql_lock() The Big QEMU Lock (BQL) has many names and they are confusing. The actual QemuMutex variable is called qemu_global_mutex but it's commonly referred to as the BQL in discussions and some code comments. The locking APIs, however, are called qemu_mutex_lock_iothread() and qemu_mutex_unlock_iothread(). The "iothread" name is historic and comes from when the main thread was split into into KVM vcpu threads and the "iothread" (now called the main loop thread). I have contributed to the confusion myself by introducing a separate --object iothread, a separate concept unrelated to the BQL. The "iothread" name is no longer appropriate for the BQL. Rename the locking APIs to: - void bql_lock(void) - void bql_unlock(void) - bool bql_locked(void) There are more APIs with "iothread" in their names. Subsequent patches will rename them. There are also comments and documentation that will be updated in later patches. Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> Reviewed-by: Paul Durrant <paul@xen.org> Acked-by: Fabiano Rosas <farosas@suse.de> Acked-by: David Woodhouse <dwmw@amazon.co.uk> Reviewed-by: Cédric Le Goater <clg@kaod.org> Acked-by: Peter Xu <peterx@redhat.com> Acked-by: Eric Farman <farman@linux.ibm.com> Reviewed-by: Harsh Prateek Bora <harshpb@linux.ibm.com> Acked-by: Hyman Huang <yong.huang@smartx.com> Reviewed-by: Akihiko Odaki <akihiko.odaki@daynix.com> Message-id: 20240102153529.486531-2-stefanha@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2024-01-02 18:35:25 +03:00
bql_unlock();
}
/*
* global_dirty_log_sync
* 1. sync dirty log from kvm
* 2. stop dirty tracking if needed.
*/
static void global_dirty_log_sync(unsigned int flag, bool one_shot)
{
system/cpus: rename qemu_mutex_lock_iothread() to bql_lock() The Big QEMU Lock (BQL) has many names and they are confusing. The actual QemuMutex variable is called qemu_global_mutex but it's commonly referred to as the BQL in discussions and some code comments. The locking APIs, however, are called qemu_mutex_lock_iothread() and qemu_mutex_unlock_iothread(). The "iothread" name is historic and comes from when the main thread was split into into KVM vcpu threads and the "iothread" (now called the main loop thread). I have contributed to the confusion myself by introducing a separate --object iothread, a separate concept unrelated to the BQL. The "iothread" name is no longer appropriate for the BQL. Rename the locking APIs to: - void bql_lock(void) - void bql_unlock(void) - bool bql_locked(void) There are more APIs with "iothread" in their names. Subsequent patches will rename them. There are also comments and documentation that will be updated in later patches. Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> Reviewed-by: Paul Durrant <paul@xen.org> Acked-by: Fabiano Rosas <farosas@suse.de> Acked-by: David Woodhouse <dwmw@amazon.co.uk> Reviewed-by: Cédric Le Goater <clg@kaod.org> Acked-by: Peter Xu <peterx@redhat.com> Acked-by: Eric Farman <farman@linux.ibm.com> Reviewed-by: Harsh Prateek Bora <harshpb@linux.ibm.com> Acked-by: Hyman Huang <yong.huang@smartx.com> Reviewed-by: Akihiko Odaki <akihiko.odaki@daynix.com> Message-id: 20240102153529.486531-2-stefanha@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2024-01-02 18:35:25 +03:00
bql_lock();
memory_global_dirty_log_sync(false);
if (one_shot) {
memory_global_dirty_log_stop(flag);
}
system/cpus: rename qemu_mutex_lock_iothread() to bql_lock() The Big QEMU Lock (BQL) has many names and they are confusing. The actual QemuMutex variable is called qemu_global_mutex but it's commonly referred to as the BQL in discussions and some code comments. The locking APIs, however, are called qemu_mutex_lock_iothread() and qemu_mutex_unlock_iothread(). The "iothread" name is historic and comes from when the main thread was split into into KVM vcpu threads and the "iothread" (now called the main loop thread). I have contributed to the confusion myself by introducing a separate --object iothread, a separate concept unrelated to the BQL. The "iothread" name is no longer appropriate for the BQL. Rename the locking APIs to: - void bql_lock(void) - void bql_unlock(void) - bool bql_locked(void) There are more APIs with "iothread" in their names. Subsequent patches will rename them. There are also comments and documentation that will be updated in later patches. Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> Reviewed-by: Paul Durrant <paul@xen.org> Acked-by: Fabiano Rosas <farosas@suse.de> Acked-by: David Woodhouse <dwmw@amazon.co.uk> Reviewed-by: Cédric Le Goater <clg@kaod.org> Acked-by: Peter Xu <peterx@redhat.com> Acked-by: Eric Farman <farman@linux.ibm.com> Reviewed-by: Harsh Prateek Bora <harshpb@linux.ibm.com> Acked-by: Hyman Huang <yong.huang@smartx.com> Reviewed-by: Akihiko Odaki <akihiko.odaki@daynix.com> Message-id: 20240102153529.486531-2-stefanha@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2024-01-02 18:35:25 +03:00
bql_unlock();
}
static DirtyPageRecord *vcpu_dirty_stat_alloc(VcpuStat *stat)
{
CPUState *cpu;
int nvcpu = 0;
CPU_FOREACH(cpu) {
nvcpu++;
}
stat->nvcpu = nvcpu;
stat->rates = g_new0(DirtyRateVcpu, nvcpu);
return g_new0(DirtyPageRecord, nvcpu);
}
static void vcpu_dirty_stat_collect(DirtyPageRecord *records,
bool start)
{
CPUState *cpu;
CPU_FOREACH(cpu) {
record_dirtypages(records, cpu, start);
}
}
int64_t vcpu_calculate_dirtyrate(int64_t calc_time_ms,
VcpuStat *stat,
unsigned int flag,
bool one_shot)
{
DirtyPageRecord *records;
int64_t init_time_ms;
int64_t duration;
int64_t dirtyrate;
int i = 0;
unsigned int gen_id;
retry:
init_time_ms = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
WITH_QEMU_LOCK_GUARD(&qemu_cpu_list_lock) {
gen_id = cpu_list_generation_id_get();
records = vcpu_dirty_stat_alloc(stat);
vcpu_dirty_stat_collect(records, true);
}
duration = dirty_stat_wait(calc_time_ms, init_time_ms);
global_dirty_log_sync(flag, one_shot);
WITH_QEMU_LOCK_GUARD(&qemu_cpu_list_lock) {
if (gen_id != cpu_list_generation_id_get()) {
g_free(records);
g_free(stat->rates);
cpu_list_unlock();
goto retry;
}
vcpu_dirty_stat_collect(records, false);
}
for (i = 0; i < stat->nvcpu; i++) {
dirtyrate = do_calculate_dirtyrate(records[i], duration);
stat->rates[i].id = i;
stat->rates[i].dirty_rate = dirtyrate;
trace_dirtyrate_do_calculate_vcpu(i, dirtyrate);
}
g_free(records);
return duration;
}
migration/calc-dirty-rate: millisecond-granularity period This patch allows to measure dirty page rate for sub-second intervals of time. An optional argument is introduced -- calc-time-unit. For example: {"execute": "calc-dirty-rate", "arguments": {"calc-time": 500, "calc-time-unit": "millisecond"} } Millisecond granularity allows to make predictions whether migration will succeed or not. To do this, calculate dirty rate with calc-time set to max allowed downtime (e.g. 300ms), convert measured rate into volume of dirtied memory, and divide by network throughput. If the value is lower than max allowed downtime, then migration will converge. Measurement results for single thread randomly writing to a 1/4/24GiB memory region: +----------------+-----------------------------------------------+ | calc-time | dirty rate MiB/s | | (milliseconds) +----------------+---------------+--------------+ | | theoretical | page-sampling | dirty-bitmap | | | (at 3M wr/sec) | | | +----------------+----------------+---------------+--------------+ | 1GiB | +----------------+----------------+---------------+--------------+ | 100 | 6996 | 7100 | 3192 | | 200 | 4606 | 4660 | 2655 | | 300 | 3305 | 3280 | 2371 | | 400 | 2534 | 2525 | 2154 | | 500 | 2041 | 2044 | 1871 | | 750 | 1365 | 1341 | 1358 | | 1000 | 1024 | 1052 | 1025 | | 1500 | 683 | 678 | 684 | | 2000 | 512 | 507 | 513 | +----------------+----------------+---------------+--------------+ | 4GiB | +----------------+----------------+---------------+--------------+ | 100 | 10232 | 8880 | 4070 | | 200 | 8954 | 8049 | 3195 | | 300 | 7889 | 7193 | 2881 | | 400 | 6996 | 6530 | 2700 | | 500 | 6245 | 5772 | 2312 | | 750 | 4829 | 4586 | 2465 | | 1000 | 3865 | 3780 | 2178 | | 1500 | 2694 | 2633 | 2004 | | 2000 | 2041 | 2031 | 1789 | +----------------+----------------+---------------+--------------+ | 24GiB | +----------------+----------------+---------------+--------------+ | 100 | 11495 | 8640 | 5597 | | 200 | 11226 | 8616 | 3527 | | 300 | 10965 | 8386 | 2355 | | 400 | 10713 | 8370 | 2179 | | 500 | 10469 | 8196 | 2098 | | 750 | 9890 | 7885 | 2556 | | 1000 | 9354 | 7506 | 2084 | | 1500 | 8397 | 6944 | 2075 | | 2000 | 7574 | 6402 | 2062 | +----------------+----------------+---------------+--------------+ Theoretical values are computed according to the following formula: size * (1 - (1-(4096/size))^(time*wps)) / (time * 2^20), where size is in bytes, time is in seconds, and wps is number of writes per second. Signed-off-by: Andrei Gudkov <gudkov.andrei@huawei.com> Reviewed-by: Hyman Huang <yong.huang@smartx.com> Message-Id: <d802e6b8053eb60fbec1a784cf86f67d9528e0a8.1693895970.git.gudkov.andrei@huawei.com> Signed-off-by: Hyman Huang <yong.huang@smartx.com>
2023-09-05 10:05:43 +03:00
static bool is_calc_time_valid(int64_t msec)
{
migration/calc-dirty-rate: millisecond-granularity period This patch allows to measure dirty page rate for sub-second intervals of time. An optional argument is introduced -- calc-time-unit. For example: {"execute": "calc-dirty-rate", "arguments": {"calc-time": 500, "calc-time-unit": "millisecond"} } Millisecond granularity allows to make predictions whether migration will succeed or not. To do this, calculate dirty rate with calc-time set to max allowed downtime (e.g. 300ms), convert measured rate into volume of dirtied memory, and divide by network throughput. If the value is lower than max allowed downtime, then migration will converge. Measurement results for single thread randomly writing to a 1/4/24GiB memory region: +----------------+-----------------------------------------------+ | calc-time | dirty rate MiB/s | | (milliseconds) +----------------+---------------+--------------+ | | theoretical | page-sampling | dirty-bitmap | | | (at 3M wr/sec) | | | +----------------+----------------+---------------+--------------+ | 1GiB | +----------------+----------------+---------------+--------------+ | 100 | 6996 | 7100 | 3192 | | 200 | 4606 | 4660 | 2655 | | 300 | 3305 | 3280 | 2371 | | 400 | 2534 | 2525 | 2154 | | 500 | 2041 | 2044 | 1871 | | 750 | 1365 | 1341 | 1358 | | 1000 | 1024 | 1052 | 1025 | | 1500 | 683 | 678 | 684 | | 2000 | 512 | 507 | 513 | +----------------+----------------+---------------+--------------+ | 4GiB | +----------------+----------------+---------------+--------------+ | 100 | 10232 | 8880 | 4070 | | 200 | 8954 | 8049 | 3195 | | 300 | 7889 | 7193 | 2881 | | 400 | 6996 | 6530 | 2700 | | 500 | 6245 | 5772 | 2312 | | 750 | 4829 | 4586 | 2465 | | 1000 | 3865 | 3780 | 2178 | | 1500 | 2694 | 2633 | 2004 | | 2000 | 2041 | 2031 | 1789 | +----------------+----------------+---------------+--------------+ | 24GiB | +----------------+----------------+---------------+--------------+ | 100 | 11495 | 8640 | 5597 | | 200 | 11226 | 8616 | 3527 | | 300 | 10965 | 8386 | 2355 | | 400 | 10713 | 8370 | 2179 | | 500 | 10469 | 8196 | 2098 | | 750 | 9890 | 7885 | 2556 | | 1000 | 9354 | 7506 | 2084 | | 1500 | 8397 | 6944 | 2075 | | 2000 | 7574 | 6402 | 2062 | +----------------+----------------+---------------+--------------+ Theoretical values are computed according to the following formula: size * (1 - (1-(4096/size))^(time*wps)) / (time * 2^20), where size is in bytes, time is in seconds, and wps is number of writes per second. Signed-off-by: Andrei Gudkov <gudkov.andrei@huawei.com> Reviewed-by: Hyman Huang <yong.huang@smartx.com> Message-Id: <d802e6b8053eb60fbec1a784cf86f67d9528e0a8.1693895970.git.gudkov.andrei@huawei.com> Signed-off-by: Hyman Huang <yong.huang@smartx.com>
2023-09-05 10:05:43 +03:00
if ((msec < MIN_CALC_TIME_MS) || (msec > MAX_CALC_TIME_MS)) {
return false;
}
return true;
}
static bool is_sample_pages_valid(int64_t pages)
{
return pages >= MIN_SAMPLE_PAGE_COUNT &&
pages <= MAX_SAMPLE_PAGE_COUNT;
}
static int dirtyrate_set_state(int *state, int old_state, int new_state)
{
assert(new_state < DIRTY_RATE_STATUS__MAX);
trace_dirtyrate_set_state(DirtyRateStatus_str(new_state));
if (qatomic_cmpxchg(state, old_state, new_state) == old_state) {
return 0;
} else {
return -1;
}
}
migration/calc-dirty-rate: millisecond-granularity period This patch allows to measure dirty page rate for sub-second intervals of time. An optional argument is introduced -- calc-time-unit. For example: {"execute": "calc-dirty-rate", "arguments": {"calc-time": 500, "calc-time-unit": "millisecond"} } Millisecond granularity allows to make predictions whether migration will succeed or not. To do this, calculate dirty rate with calc-time set to max allowed downtime (e.g. 300ms), convert measured rate into volume of dirtied memory, and divide by network throughput. If the value is lower than max allowed downtime, then migration will converge. Measurement results for single thread randomly writing to a 1/4/24GiB memory region: +----------------+-----------------------------------------------+ | calc-time | dirty rate MiB/s | | (milliseconds) +----------------+---------------+--------------+ | | theoretical | page-sampling | dirty-bitmap | | | (at 3M wr/sec) | | | +----------------+----------------+---------------+--------------+ | 1GiB | +----------------+----------------+---------------+--------------+ | 100 | 6996 | 7100 | 3192 | | 200 | 4606 | 4660 | 2655 | | 300 | 3305 | 3280 | 2371 | | 400 | 2534 | 2525 | 2154 | | 500 | 2041 | 2044 | 1871 | | 750 | 1365 | 1341 | 1358 | | 1000 | 1024 | 1052 | 1025 | | 1500 | 683 | 678 | 684 | | 2000 | 512 | 507 | 513 | +----------------+----------------+---------------+--------------+ | 4GiB | +----------------+----------------+---------------+--------------+ | 100 | 10232 | 8880 | 4070 | | 200 | 8954 | 8049 | 3195 | | 300 | 7889 | 7193 | 2881 | | 400 | 6996 | 6530 | 2700 | | 500 | 6245 | 5772 | 2312 | | 750 | 4829 | 4586 | 2465 | | 1000 | 3865 | 3780 | 2178 | | 1500 | 2694 | 2633 | 2004 | | 2000 | 2041 | 2031 | 1789 | +----------------+----------------+---------------+--------------+ | 24GiB | +----------------+----------------+---------------+--------------+ | 100 | 11495 | 8640 | 5597 | | 200 | 11226 | 8616 | 3527 | | 300 | 10965 | 8386 | 2355 | | 400 | 10713 | 8370 | 2179 | | 500 | 10469 | 8196 | 2098 | | 750 | 9890 | 7885 | 2556 | | 1000 | 9354 | 7506 | 2084 | | 1500 | 8397 | 6944 | 2075 | | 2000 | 7574 | 6402 | 2062 | +----------------+----------------+---------------+--------------+ Theoretical values are computed according to the following formula: size * (1 - (1-(4096/size))^(time*wps)) / (time * 2^20), where size is in bytes, time is in seconds, and wps is number of writes per second. Signed-off-by: Andrei Gudkov <gudkov.andrei@huawei.com> Reviewed-by: Hyman Huang <yong.huang@smartx.com> Message-Id: <d802e6b8053eb60fbec1a784cf86f67d9528e0a8.1693895970.git.gudkov.andrei@huawei.com> Signed-off-by: Hyman Huang <yong.huang@smartx.com>
2023-09-05 10:05:43 +03:00
/* Decimal power of given time unit relative to one second */
static int time_unit_to_power(TimeUnit time_unit)
{
switch (time_unit) {
case TIME_UNIT_SECOND:
return 0;
case TIME_UNIT_MILLISECOND:
return -3;
default:
assert(false); /* unreachable */
return 0;
}
}
static int64_t convert_time_unit(int64_t value, TimeUnit unit_from,
TimeUnit unit_to)
{
int power = time_unit_to_power(unit_from) -
time_unit_to_power(unit_to);
while (power < 0) {
value /= 10;
power += 1;
}
while (power > 0) {
value *= 10;
power -= 1;
}
return value;
}
static struct DirtyRateInfo *
query_dirty_rate_info(TimeUnit calc_time_unit)
{
int i;
int64_t dirty_rate = DirtyStat.dirty_rate;
struct DirtyRateInfo *info = g_new0(DirtyRateInfo, 1);
DirtyRateVcpuList *head = NULL, **tail = &head;
info->status = CalculatingState;
info->start_time = DirtyStat.start_time;
migration/calc-dirty-rate: millisecond-granularity period This patch allows to measure dirty page rate for sub-second intervals of time. An optional argument is introduced -- calc-time-unit. For example: {"execute": "calc-dirty-rate", "arguments": {"calc-time": 500, "calc-time-unit": "millisecond"} } Millisecond granularity allows to make predictions whether migration will succeed or not. To do this, calculate dirty rate with calc-time set to max allowed downtime (e.g. 300ms), convert measured rate into volume of dirtied memory, and divide by network throughput. If the value is lower than max allowed downtime, then migration will converge. Measurement results for single thread randomly writing to a 1/4/24GiB memory region: +----------------+-----------------------------------------------+ | calc-time | dirty rate MiB/s | | (milliseconds) +----------------+---------------+--------------+ | | theoretical | page-sampling | dirty-bitmap | | | (at 3M wr/sec) | | | +----------------+----------------+---------------+--------------+ | 1GiB | +----------------+----------------+---------------+--------------+ | 100 | 6996 | 7100 | 3192 | | 200 | 4606 | 4660 | 2655 | | 300 | 3305 | 3280 | 2371 | | 400 | 2534 | 2525 | 2154 | | 500 | 2041 | 2044 | 1871 | | 750 | 1365 | 1341 | 1358 | | 1000 | 1024 | 1052 | 1025 | | 1500 | 683 | 678 | 684 | | 2000 | 512 | 507 | 513 | +----------------+----------------+---------------+--------------+ | 4GiB | +----------------+----------------+---------------+--------------+ | 100 | 10232 | 8880 | 4070 | | 200 | 8954 | 8049 | 3195 | | 300 | 7889 | 7193 | 2881 | | 400 | 6996 | 6530 | 2700 | | 500 | 6245 | 5772 | 2312 | | 750 | 4829 | 4586 | 2465 | | 1000 | 3865 | 3780 | 2178 | | 1500 | 2694 | 2633 | 2004 | | 2000 | 2041 | 2031 | 1789 | +----------------+----------------+---------------+--------------+ | 24GiB | +----------------+----------------+---------------+--------------+ | 100 | 11495 | 8640 | 5597 | | 200 | 11226 | 8616 | 3527 | | 300 | 10965 | 8386 | 2355 | | 400 | 10713 | 8370 | 2179 | | 500 | 10469 | 8196 | 2098 | | 750 | 9890 | 7885 | 2556 | | 1000 | 9354 | 7506 | 2084 | | 1500 | 8397 | 6944 | 2075 | | 2000 | 7574 | 6402 | 2062 | +----------------+----------------+---------------+--------------+ Theoretical values are computed according to the following formula: size * (1 - (1-(4096/size))^(time*wps)) / (time * 2^20), where size is in bytes, time is in seconds, and wps is number of writes per second. Signed-off-by: Andrei Gudkov <gudkov.andrei@huawei.com> Reviewed-by: Hyman Huang <yong.huang@smartx.com> Message-Id: <d802e6b8053eb60fbec1a784cf86f67d9528e0a8.1693895970.git.gudkov.andrei@huawei.com> Signed-off-by: Hyman Huang <yong.huang@smartx.com>
2023-09-05 10:05:43 +03:00
info->calc_time = convert_time_unit(DirtyStat.calc_time_ms,
TIME_UNIT_MILLISECOND,
calc_time_unit);
info->calc_time_unit = calc_time_unit;
info->sample_pages = DirtyStat.sample_pages;
info->mode = dirtyrate_mode;
if (qatomic_read(&CalculatingState) == DIRTY_RATE_STATUS_MEASURED) {
info->has_dirty_rate = true;
info->dirty_rate = dirty_rate;
if (dirtyrate_mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) {
/*
* set sample_pages with 0 to indicate page sampling
* isn't enabled
**/
info->sample_pages = 0;
info->has_vcpu_dirty_rate = true;
for (i = 0; i < DirtyStat.dirty_ring.nvcpu; i++) {
DirtyRateVcpu *rate = g_new0(DirtyRateVcpu, 1);
rate->id = DirtyStat.dirty_ring.rates[i].id;
rate->dirty_rate = DirtyStat.dirty_ring.rates[i].dirty_rate;
QAPI_LIST_APPEND(tail, rate);
}
info->vcpu_dirty_rate = head;
}
if (dirtyrate_mode == DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP) {
info->sample_pages = 0;
}
}
trace_query_dirty_rate_info(DirtyRateStatus_str(CalculatingState));
return info;
}
static void init_dirtyrate_stat(struct DirtyRateConfig config)
{
DirtyStat.dirty_rate = -1;
DirtyStat.start_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) / 1000;
migration/calc-dirty-rate: millisecond-granularity period This patch allows to measure dirty page rate for sub-second intervals of time. An optional argument is introduced -- calc-time-unit. For example: {"execute": "calc-dirty-rate", "arguments": {"calc-time": 500, "calc-time-unit": "millisecond"} } Millisecond granularity allows to make predictions whether migration will succeed or not. To do this, calculate dirty rate with calc-time set to max allowed downtime (e.g. 300ms), convert measured rate into volume of dirtied memory, and divide by network throughput. If the value is lower than max allowed downtime, then migration will converge. Measurement results for single thread randomly writing to a 1/4/24GiB memory region: +----------------+-----------------------------------------------+ | calc-time | dirty rate MiB/s | | (milliseconds) +----------------+---------------+--------------+ | | theoretical | page-sampling | dirty-bitmap | | | (at 3M wr/sec) | | | +----------------+----------------+---------------+--------------+ | 1GiB | +----------------+----------------+---------------+--------------+ | 100 | 6996 | 7100 | 3192 | | 200 | 4606 | 4660 | 2655 | | 300 | 3305 | 3280 | 2371 | | 400 | 2534 | 2525 | 2154 | | 500 | 2041 | 2044 | 1871 | | 750 | 1365 | 1341 | 1358 | | 1000 | 1024 | 1052 | 1025 | | 1500 | 683 | 678 | 684 | | 2000 | 512 | 507 | 513 | +----------------+----------------+---------------+--------------+ | 4GiB | +----------------+----------------+---------------+--------------+ | 100 | 10232 | 8880 | 4070 | | 200 | 8954 | 8049 | 3195 | | 300 | 7889 | 7193 | 2881 | | 400 | 6996 | 6530 | 2700 | | 500 | 6245 | 5772 | 2312 | | 750 | 4829 | 4586 | 2465 | | 1000 | 3865 | 3780 | 2178 | | 1500 | 2694 | 2633 | 2004 | | 2000 | 2041 | 2031 | 1789 | +----------------+----------------+---------------+--------------+ | 24GiB | +----------------+----------------+---------------+--------------+ | 100 | 11495 | 8640 | 5597 | | 200 | 11226 | 8616 | 3527 | | 300 | 10965 | 8386 | 2355 | | 400 | 10713 | 8370 | 2179 | | 500 | 10469 | 8196 | 2098 | | 750 | 9890 | 7885 | 2556 | | 1000 | 9354 | 7506 | 2084 | | 1500 | 8397 | 6944 | 2075 | | 2000 | 7574 | 6402 | 2062 | +----------------+----------------+---------------+--------------+ Theoretical values are computed according to the following formula: size * (1 - (1-(4096/size))^(time*wps)) / (time * 2^20), where size is in bytes, time is in seconds, and wps is number of writes per second. Signed-off-by: Andrei Gudkov <gudkov.andrei@huawei.com> Reviewed-by: Hyman Huang <yong.huang@smartx.com> Message-Id: <d802e6b8053eb60fbec1a784cf86f67d9528e0a8.1693895970.git.gudkov.andrei@huawei.com> Signed-off-by: Hyman Huang <yong.huang@smartx.com>
2023-09-05 10:05:43 +03:00
DirtyStat.calc_time_ms = config.calc_time_ms;
DirtyStat.sample_pages = config.sample_pages_per_gigabytes;
switch (config.mode) {
case DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING:
DirtyStat.page_sampling.total_dirty_samples = 0;
DirtyStat.page_sampling.total_sample_count = 0;
DirtyStat.page_sampling.total_block_mem_MB = 0;
break;
case DIRTY_RATE_MEASURE_MODE_DIRTY_RING:
DirtyStat.dirty_ring.nvcpu = -1;
DirtyStat.dirty_ring.rates = NULL;
break;
default:
break;
}
}
static void cleanup_dirtyrate_stat(struct DirtyRateConfig config)
{
/* last calc-dirty-rate qmp use dirty ring mode */
if (dirtyrate_mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) {
free(DirtyStat.dirty_ring.rates);
DirtyStat.dirty_ring.rates = NULL;
}
}
static void update_dirtyrate_stat(struct RamblockDirtyInfo *info)
{
DirtyStat.page_sampling.total_dirty_samples += info->sample_dirty_count;
DirtyStat.page_sampling.total_sample_count += info->sample_pages_count;
/* size of total pages in MB */
DirtyStat.page_sampling.total_block_mem_MB +=
qemu_target_pages_to_MiB(info->ramblock_pages);
}
static void update_dirtyrate(uint64_t msec)
{
uint64_t dirtyrate;
uint64_t total_dirty_samples = DirtyStat.page_sampling.total_dirty_samples;
uint64_t total_sample_count = DirtyStat.page_sampling.total_sample_count;
uint64_t total_block_mem_MB = DirtyStat.page_sampling.total_block_mem_MB;
dirtyrate = total_dirty_samples * total_block_mem_MB *
1000 / (total_sample_count * msec);
DirtyStat.dirty_rate = dirtyrate;
}
/*
* Compute hash of a single page of size TARGET_PAGE_SIZE.
*/
static uint32_t compute_page_hash(void *ptr)
{
size_t page_size = qemu_target_page_size();
uint32_t i;
uint64_t v1, v2, v3, v4;
uint64_t res;
const uint64_t *p = ptr;
v1 = QEMU_XXHASH_SEED + XXH_PRIME64_1 + XXH_PRIME64_2;
v2 = QEMU_XXHASH_SEED + XXH_PRIME64_2;
v3 = QEMU_XXHASH_SEED + 0;
v4 = QEMU_XXHASH_SEED - XXH_PRIME64_1;
for (i = 0; i < page_size / 8; i += 4) {
v1 = XXH64_round(v1, p[i + 0]);
v2 = XXH64_round(v2, p[i + 1]);
v3 = XXH64_round(v3, p[i + 2]);
v4 = XXH64_round(v4, p[i + 3]);
}
res = XXH64_mergerounds(v1, v2, v3, v4);
res += page_size;
res = XXH64_avalanche(res);
return (uint32_t)(res & UINT32_MAX);
}
/*
* get hash result for the sampled memory with length of TARGET_PAGE_SIZE
* in ramblock, which starts from ramblock base address.
*/
static uint32_t get_ramblock_vfn_hash(struct RamblockDirtyInfo *info,
uint64_t vfn)
{
uint32_t hash;
hash = compute_page_hash(info->ramblock_addr +
vfn * qemu_target_page_size());
trace_get_ramblock_vfn_hash(info->idstr, vfn, hash);
return hash;
}
static bool save_ramblock_hash(struct RamblockDirtyInfo *info)
{
unsigned int sample_pages_count;
int i;
GRand *rand;
sample_pages_count = info->sample_pages_count;
/* ramblock size less than one page, return success to skip this ramblock */
if (unlikely(info->ramblock_pages == 0 || sample_pages_count == 0)) {
return true;
}
info->hash_result = g_try_malloc0_n(sample_pages_count,
sizeof(uint32_t));
if (!info->hash_result) {
return false;
}
info->sample_page_vfn = g_try_malloc0_n(sample_pages_count,
sizeof(uint64_t));
if (!info->sample_page_vfn) {
g_free(info->hash_result);
return false;
}
rand = g_rand_new();
for (i = 0; i < sample_pages_count; i++) {
info->sample_page_vfn[i] = g_rand_int_range(rand, 0,
info->ramblock_pages - 1);
info->hash_result[i] = get_ramblock_vfn_hash(info,
info->sample_page_vfn[i]);
}
g_rand_free(rand);
return true;
}
static void get_ramblock_dirty_info(RAMBlock *block,
struct RamblockDirtyInfo *info,
struct DirtyRateConfig *config)
{
uint64_t sample_pages_per_gigabytes = config->sample_pages_per_gigabytes;
/* Right shift 30 bits to calc ramblock size in GB */
info->sample_pages_count = (qemu_ram_get_used_length(block) *
sample_pages_per_gigabytes) >> 30;
/* Right shift TARGET_PAGE_BITS to calc page count */
info->ramblock_pages = qemu_ram_get_used_length(block) >>
qemu_target_page_bits();
info->ramblock_addr = qemu_ram_get_host_addr(block);
strcpy(info->idstr, qemu_ram_get_idstr(block));
}
static void free_ramblock_dirty_info(struct RamblockDirtyInfo *infos, int count)
{
int i;
if (!infos) {
return;
}
for (i = 0; i < count; i++) {
g_free(infos[i].sample_page_vfn);
g_free(infos[i].hash_result);
}
g_free(infos);
}
static bool skip_sample_ramblock(RAMBlock *block)
{
/*
* Sample only blocks larger than MIN_RAMBLOCK_SIZE.
*/
if (qemu_ram_get_used_length(block) < (MIN_RAMBLOCK_SIZE << 10)) {
trace_skip_sample_ramblock(block->idstr,
qemu_ram_get_used_length(block));
return true;
}
return false;
}
static bool record_ramblock_hash_info(struct RamblockDirtyInfo **block_dinfo,
struct DirtyRateConfig config,
int *block_count)
{
struct RamblockDirtyInfo *info = NULL;
struct RamblockDirtyInfo *dinfo = NULL;
RAMBlock *block = NULL;
int total_count = 0;
int index = 0;
bool ret = false;
RAMBLOCK_FOREACH_MIGRATABLE(block) {
if (skip_sample_ramblock(block)) {
continue;
}
total_count++;
}
dinfo = g_try_malloc0_n(total_count, sizeof(struct RamblockDirtyInfo));
if (dinfo == NULL) {
goto out;
}
RAMBLOCK_FOREACH_MIGRATABLE(block) {
if (skip_sample_ramblock(block)) {
continue;
}
if (index >= total_count) {
break;
}
info = &dinfo[index];
get_ramblock_dirty_info(block, info, &config);
if (!save_ramblock_hash(info)) {
goto out;
}
index++;
}
ret = true;
out:
*block_count = index;
*block_dinfo = dinfo;
return ret;
}
static void calc_page_dirty_rate(struct RamblockDirtyInfo *info)
{
uint32_t hash;
int i;
for (i = 0; i < info->sample_pages_count; i++) {
hash = get_ramblock_vfn_hash(info, info->sample_page_vfn[i]);
if (hash != info->hash_result[i]) {
trace_calc_page_dirty_rate(info->idstr, hash, info->hash_result[i]);
info->sample_dirty_count++;
}
}
}
static struct RamblockDirtyInfo *
find_block_matched(RAMBlock *block, int count,
struct RamblockDirtyInfo *infos)
{
int i;
for (i = 0; i < count; i++) {
if (!strcmp(infos[i].idstr, qemu_ram_get_idstr(block))) {
break;
}
}
if (i == count) {
return NULL;
}
if (infos[i].ramblock_addr != qemu_ram_get_host_addr(block) ||
infos[i].ramblock_pages !=
(qemu_ram_get_used_length(block) >> qemu_target_page_bits())) {
trace_find_page_matched(block->idstr);
return NULL;
}
return &infos[i];
}
static bool compare_page_hash_info(struct RamblockDirtyInfo *info,
int block_count)
{
struct RamblockDirtyInfo *block_dinfo = NULL;
RAMBlock *block = NULL;
RAMBLOCK_FOREACH_MIGRATABLE(block) {
if (skip_sample_ramblock(block)) {
continue;
}
block_dinfo = find_block_matched(block, block_count, info);
if (block_dinfo == NULL) {
continue;
}
calc_page_dirty_rate(block_dinfo);
update_dirtyrate_stat(block_dinfo);
}
if (DirtyStat.page_sampling.total_sample_count == 0) {
return false;
}
return true;
}
static inline void record_dirtypages_bitmap(DirtyPageRecord *dirty_pages,
bool start)
{
if (start) {
dirty_pages->start_pages = total_dirty_pages;
} else {
dirty_pages->end_pages = total_dirty_pages;
}
}
static inline void dirtyrate_manual_reset_protect(void)
{
RAMBlock *block = NULL;
WITH_RCU_READ_LOCK_GUARD() {
RAMBLOCK_FOREACH_MIGRATABLE(block) {
memory_region_clear_dirty_bitmap(block->mr, 0,
block->used_length);
}
}
}
static void calculate_dirtyrate_dirty_bitmap(struct DirtyRateConfig config)
{
int64_t start_time;
DirtyPageRecord dirty_pages;
system/cpus: rename qemu_mutex_lock_iothread() to bql_lock() The Big QEMU Lock (BQL) has many names and they are confusing. The actual QemuMutex variable is called qemu_global_mutex but it's commonly referred to as the BQL in discussions and some code comments. The locking APIs, however, are called qemu_mutex_lock_iothread() and qemu_mutex_unlock_iothread(). The "iothread" name is historic and comes from when the main thread was split into into KVM vcpu threads and the "iothread" (now called the main loop thread). I have contributed to the confusion myself by introducing a separate --object iothread, a separate concept unrelated to the BQL. The "iothread" name is no longer appropriate for the BQL. Rename the locking APIs to: - void bql_lock(void) - void bql_unlock(void) - bool bql_locked(void) There are more APIs with "iothread" in their names. Subsequent patches will rename them. There are also comments and documentation that will be updated in later patches. Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> Reviewed-by: Paul Durrant <paul@xen.org> Acked-by: Fabiano Rosas <farosas@suse.de> Acked-by: David Woodhouse <dwmw@amazon.co.uk> Reviewed-by: Cédric Le Goater <clg@kaod.org> Acked-by: Peter Xu <peterx@redhat.com> Acked-by: Eric Farman <farman@linux.ibm.com> Reviewed-by: Harsh Prateek Bora <harshpb@linux.ibm.com> Acked-by: Hyman Huang <yong.huang@smartx.com> Reviewed-by: Akihiko Odaki <akihiko.odaki@daynix.com> Message-id: 20240102153529.486531-2-stefanha@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2024-01-02 18:35:25 +03:00
bql_lock();
memory_global_dirty_log_start(GLOBAL_DIRTY_DIRTY_RATE);
/*
* 1'round of log sync may return all 1 bits with
* KVM_DIRTY_LOG_INITIALLY_SET enable
* skip it unconditionally and start dirty tracking
* from 2'round of log sync
*/
memory_global_dirty_log_sync(false);
/*
* reset page protect manually and unconditionally.
* this make sure kvm dirty log be cleared if
* KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE cap is enabled.
*/
dirtyrate_manual_reset_protect();
system/cpus: rename qemu_mutex_lock_iothread() to bql_lock() The Big QEMU Lock (BQL) has many names and they are confusing. The actual QemuMutex variable is called qemu_global_mutex but it's commonly referred to as the BQL in discussions and some code comments. The locking APIs, however, are called qemu_mutex_lock_iothread() and qemu_mutex_unlock_iothread(). The "iothread" name is historic and comes from when the main thread was split into into KVM vcpu threads and the "iothread" (now called the main loop thread). I have contributed to the confusion myself by introducing a separate --object iothread, a separate concept unrelated to the BQL. The "iothread" name is no longer appropriate for the BQL. Rename the locking APIs to: - void bql_lock(void) - void bql_unlock(void) - bool bql_locked(void) There are more APIs with "iothread" in their names. Subsequent patches will rename them. There are also comments and documentation that will be updated in later patches. Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> Reviewed-by: Paul Durrant <paul@xen.org> Acked-by: Fabiano Rosas <farosas@suse.de> Acked-by: David Woodhouse <dwmw@amazon.co.uk> Reviewed-by: Cédric Le Goater <clg@kaod.org> Acked-by: Peter Xu <peterx@redhat.com> Acked-by: Eric Farman <farman@linux.ibm.com> Reviewed-by: Harsh Prateek Bora <harshpb@linux.ibm.com> Acked-by: Hyman Huang <yong.huang@smartx.com> Reviewed-by: Akihiko Odaki <akihiko.odaki@daynix.com> Message-id: 20240102153529.486531-2-stefanha@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2024-01-02 18:35:25 +03:00
bql_unlock();
record_dirtypages_bitmap(&dirty_pages, true);
start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
DirtyStat.start_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) / 1000;
migration/calc-dirty-rate: millisecond-granularity period This patch allows to measure dirty page rate for sub-second intervals of time. An optional argument is introduced -- calc-time-unit. For example: {"execute": "calc-dirty-rate", "arguments": {"calc-time": 500, "calc-time-unit": "millisecond"} } Millisecond granularity allows to make predictions whether migration will succeed or not. To do this, calculate dirty rate with calc-time set to max allowed downtime (e.g. 300ms), convert measured rate into volume of dirtied memory, and divide by network throughput. If the value is lower than max allowed downtime, then migration will converge. Measurement results for single thread randomly writing to a 1/4/24GiB memory region: +----------------+-----------------------------------------------+ | calc-time | dirty rate MiB/s | | (milliseconds) +----------------+---------------+--------------+ | | theoretical | page-sampling | dirty-bitmap | | | (at 3M wr/sec) | | | +----------------+----------------+---------------+--------------+ | 1GiB | +----------------+----------------+---------------+--------------+ | 100 | 6996 | 7100 | 3192 | | 200 | 4606 | 4660 | 2655 | | 300 | 3305 | 3280 | 2371 | | 400 | 2534 | 2525 | 2154 | | 500 | 2041 | 2044 | 1871 | | 750 | 1365 | 1341 | 1358 | | 1000 | 1024 | 1052 | 1025 | | 1500 | 683 | 678 | 684 | | 2000 | 512 | 507 | 513 | +----------------+----------------+---------------+--------------+ | 4GiB | +----------------+----------------+---------------+--------------+ | 100 | 10232 | 8880 | 4070 | | 200 | 8954 | 8049 | 3195 | | 300 | 7889 | 7193 | 2881 | | 400 | 6996 | 6530 | 2700 | | 500 | 6245 | 5772 | 2312 | | 750 | 4829 | 4586 | 2465 | | 1000 | 3865 | 3780 | 2178 | | 1500 | 2694 | 2633 | 2004 | | 2000 | 2041 | 2031 | 1789 | +----------------+----------------+---------------+--------------+ | 24GiB | +----------------+----------------+---------------+--------------+ | 100 | 11495 | 8640 | 5597 | | 200 | 11226 | 8616 | 3527 | | 300 | 10965 | 8386 | 2355 | | 400 | 10713 | 8370 | 2179 | | 500 | 10469 | 8196 | 2098 | | 750 | 9890 | 7885 | 2556 | | 1000 | 9354 | 7506 | 2084 | | 1500 | 8397 | 6944 | 2075 | | 2000 | 7574 | 6402 | 2062 | +----------------+----------------+---------------+--------------+ Theoretical values are computed according to the following formula: size * (1 - (1-(4096/size))^(time*wps)) / (time * 2^20), where size is in bytes, time is in seconds, and wps is number of writes per second. Signed-off-by: Andrei Gudkov <gudkov.andrei@huawei.com> Reviewed-by: Hyman Huang <yong.huang@smartx.com> Message-Id: <d802e6b8053eb60fbec1a784cf86f67d9528e0a8.1693895970.git.gudkov.andrei@huawei.com> Signed-off-by: Hyman Huang <yong.huang@smartx.com>
2023-09-05 10:05:43 +03:00
DirtyStat.calc_time_ms = dirty_stat_wait(config.calc_time_ms, start_time);
/*
* do two things.
* 1. fetch dirty bitmap from kvm
* 2. stop dirty tracking
*/
global_dirty_log_sync(GLOBAL_DIRTY_DIRTY_RATE, true);
record_dirtypages_bitmap(&dirty_pages, false);
migration/calc-dirty-rate: millisecond-granularity period This patch allows to measure dirty page rate for sub-second intervals of time. An optional argument is introduced -- calc-time-unit. For example: {"execute": "calc-dirty-rate", "arguments": {"calc-time": 500, "calc-time-unit": "millisecond"} } Millisecond granularity allows to make predictions whether migration will succeed or not. To do this, calculate dirty rate with calc-time set to max allowed downtime (e.g. 300ms), convert measured rate into volume of dirtied memory, and divide by network throughput. If the value is lower than max allowed downtime, then migration will converge. Measurement results for single thread randomly writing to a 1/4/24GiB memory region: +----------------+-----------------------------------------------+ | calc-time | dirty rate MiB/s | | (milliseconds) +----------------+---------------+--------------+ | | theoretical | page-sampling | dirty-bitmap | | | (at 3M wr/sec) | | | +----------------+----------------+---------------+--------------+ | 1GiB | +----------------+----------------+---------------+--------------+ | 100 | 6996 | 7100 | 3192 | | 200 | 4606 | 4660 | 2655 | | 300 | 3305 | 3280 | 2371 | | 400 | 2534 | 2525 | 2154 | | 500 | 2041 | 2044 | 1871 | | 750 | 1365 | 1341 | 1358 | | 1000 | 1024 | 1052 | 1025 | | 1500 | 683 | 678 | 684 | | 2000 | 512 | 507 | 513 | +----------------+----------------+---------------+--------------+ | 4GiB | +----------------+----------------+---------------+--------------+ | 100 | 10232 | 8880 | 4070 | | 200 | 8954 | 8049 | 3195 | | 300 | 7889 | 7193 | 2881 | | 400 | 6996 | 6530 | 2700 | | 500 | 6245 | 5772 | 2312 | | 750 | 4829 | 4586 | 2465 | | 1000 | 3865 | 3780 | 2178 | | 1500 | 2694 | 2633 | 2004 | | 2000 | 2041 | 2031 | 1789 | +----------------+----------------+---------------+--------------+ | 24GiB | +----------------+----------------+---------------+--------------+ | 100 | 11495 | 8640 | 5597 | | 200 | 11226 | 8616 | 3527 | | 300 | 10965 | 8386 | 2355 | | 400 | 10713 | 8370 | 2179 | | 500 | 10469 | 8196 | 2098 | | 750 | 9890 | 7885 | 2556 | | 1000 | 9354 | 7506 | 2084 | | 1500 | 8397 | 6944 | 2075 | | 2000 | 7574 | 6402 | 2062 | +----------------+----------------+---------------+--------------+ Theoretical values are computed according to the following formula: size * (1 - (1-(4096/size))^(time*wps)) / (time * 2^20), where size is in bytes, time is in seconds, and wps is number of writes per second. Signed-off-by: Andrei Gudkov <gudkov.andrei@huawei.com> Reviewed-by: Hyman Huang <yong.huang@smartx.com> Message-Id: <d802e6b8053eb60fbec1a784cf86f67d9528e0a8.1693895970.git.gudkov.andrei@huawei.com> Signed-off-by: Hyman Huang <yong.huang@smartx.com>
2023-09-05 10:05:43 +03:00
DirtyStat.dirty_rate = do_calculate_dirtyrate(dirty_pages,
DirtyStat.calc_time_ms);
}
static void calculate_dirtyrate_dirty_ring(struct DirtyRateConfig config)
{
uint64_t dirtyrate = 0;
uint64_t dirtyrate_sum = 0;
int i = 0;
/* start log sync */
global_dirty_log_change(GLOBAL_DIRTY_DIRTY_RATE, true);
DirtyStat.start_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) / 1000;
/* calculate vcpu dirtyrate */
migration/calc-dirty-rate: millisecond-granularity period This patch allows to measure dirty page rate for sub-second intervals of time. An optional argument is introduced -- calc-time-unit. For example: {"execute": "calc-dirty-rate", "arguments": {"calc-time": 500, "calc-time-unit": "millisecond"} } Millisecond granularity allows to make predictions whether migration will succeed or not. To do this, calculate dirty rate with calc-time set to max allowed downtime (e.g. 300ms), convert measured rate into volume of dirtied memory, and divide by network throughput. If the value is lower than max allowed downtime, then migration will converge. Measurement results for single thread randomly writing to a 1/4/24GiB memory region: +----------------+-----------------------------------------------+ | calc-time | dirty rate MiB/s | | (milliseconds) +----------------+---------------+--------------+ | | theoretical | page-sampling | dirty-bitmap | | | (at 3M wr/sec) | | | +----------------+----------------+---------------+--------------+ | 1GiB | +----------------+----------------+---------------+--------------+ | 100 | 6996 | 7100 | 3192 | | 200 | 4606 | 4660 | 2655 | | 300 | 3305 | 3280 | 2371 | | 400 | 2534 | 2525 | 2154 | | 500 | 2041 | 2044 | 1871 | | 750 | 1365 | 1341 | 1358 | | 1000 | 1024 | 1052 | 1025 | | 1500 | 683 | 678 | 684 | | 2000 | 512 | 507 | 513 | +----------------+----------------+---------------+--------------+ | 4GiB | +----------------+----------------+---------------+--------------+ | 100 | 10232 | 8880 | 4070 | | 200 | 8954 | 8049 | 3195 | | 300 | 7889 | 7193 | 2881 | | 400 | 6996 | 6530 | 2700 | | 500 | 6245 | 5772 | 2312 | | 750 | 4829 | 4586 | 2465 | | 1000 | 3865 | 3780 | 2178 | | 1500 | 2694 | 2633 | 2004 | | 2000 | 2041 | 2031 | 1789 | +----------------+----------------+---------------+--------------+ | 24GiB | +----------------+----------------+---------------+--------------+ | 100 | 11495 | 8640 | 5597 | | 200 | 11226 | 8616 | 3527 | | 300 | 10965 | 8386 | 2355 | | 400 | 10713 | 8370 | 2179 | | 500 | 10469 | 8196 | 2098 | | 750 | 9890 | 7885 | 2556 | | 1000 | 9354 | 7506 | 2084 | | 1500 | 8397 | 6944 | 2075 | | 2000 | 7574 | 6402 | 2062 | +----------------+----------------+---------------+--------------+ Theoretical values are computed according to the following formula: size * (1 - (1-(4096/size))^(time*wps)) / (time * 2^20), where size is in bytes, time is in seconds, and wps is number of writes per second. Signed-off-by: Andrei Gudkov <gudkov.andrei@huawei.com> Reviewed-by: Hyman Huang <yong.huang@smartx.com> Message-Id: <d802e6b8053eb60fbec1a784cf86f67d9528e0a8.1693895970.git.gudkov.andrei@huawei.com> Signed-off-by: Hyman Huang <yong.huang@smartx.com>
2023-09-05 10:05:43 +03:00
DirtyStat.calc_time_ms = vcpu_calculate_dirtyrate(config.calc_time_ms,
&DirtyStat.dirty_ring,
GLOBAL_DIRTY_DIRTY_RATE,
true);
/* calculate vm dirtyrate */
for (i = 0; i < DirtyStat.dirty_ring.nvcpu; i++) {
dirtyrate = DirtyStat.dirty_ring.rates[i].dirty_rate;
DirtyStat.dirty_ring.rates[i].dirty_rate = dirtyrate;
dirtyrate_sum += dirtyrate;
}
DirtyStat.dirty_rate = dirtyrate_sum;
}
static void calculate_dirtyrate_sample_vm(struct DirtyRateConfig config)
{
struct RamblockDirtyInfo *block_dinfo = NULL;
int block_count = 0;
int64_t initial_time;
rcu_read_lock();
initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
DirtyStat.start_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) / 1000;
if (!record_ramblock_hash_info(&block_dinfo, config, &block_count)) {
goto out;
}
rcu_read_unlock();
migration/calc-dirty-rate: millisecond-granularity period This patch allows to measure dirty page rate for sub-second intervals of time. An optional argument is introduced -- calc-time-unit. For example: {"execute": "calc-dirty-rate", "arguments": {"calc-time": 500, "calc-time-unit": "millisecond"} } Millisecond granularity allows to make predictions whether migration will succeed or not. To do this, calculate dirty rate with calc-time set to max allowed downtime (e.g. 300ms), convert measured rate into volume of dirtied memory, and divide by network throughput. If the value is lower than max allowed downtime, then migration will converge. Measurement results for single thread randomly writing to a 1/4/24GiB memory region: +----------------+-----------------------------------------------+ | calc-time | dirty rate MiB/s | | (milliseconds) +----------------+---------------+--------------+ | | theoretical | page-sampling | dirty-bitmap | | | (at 3M wr/sec) | | | +----------------+----------------+---------------+--------------+ | 1GiB | +----------------+----------------+---------------+--------------+ | 100 | 6996 | 7100 | 3192 | | 200 | 4606 | 4660 | 2655 | | 300 | 3305 | 3280 | 2371 | | 400 | 2534 | 2525 | 2154 | | 500 | 2041 | 2044 | 1871 | | 750 | 1365 | 1341 | 1358 | | 1000 | 1024 | 1052 | 1025 | | 1500 | 683 | 678 | 684 | | 2000 | 512 | 507 | 513 | +----------------+----------------+---------------+--------------+ | 4GiB | +----------------+----------------+---------------+--------------+ | 100 | 10232 | 8880 | 4070 | | 200 | 8954 | 8049 | 3195 | | 300 | 7889 | 7193 | 2881 | | 400 | 6996 | 6530 | 2700 | | 500 | 6245 | 5772 | 2312 | | 750 | 4829 | 4586 | 2465 | | 1000 | 3865 | 3780 | 2178 | | 1500 | 2694 | 2633 | 2004 | | 2000 | 2041 | 2031 | 1789 | +----------------+----------------+---------------+--------------+ | 24GiB | +----------------+----------------+---------------+--------------+ | 100 | 11495 | 8640 | 5597 | | 200 | 11226 | 8616 | 3527 | | 300 | 10965 | 8386 | 2355 | | 400 | 10713 | 8370 | 2179 | | 500 | 10469 | 8196 | 2098 | | 750 | 9890 | 7885 | 2556 | | 1000 | 9354 | 7506 | 2084 | | 1500 | 8397 | 6944 | 2075 | | 2000 | 7574 | 6402 | 2062 | +----------------+----------------+---------------+--------------+ Theoretical values are computed according to the following formula: size * (1 - (1-(4096/size))^(time*wps)) / (time * 2^20), where size is in bytes, time is in seconds, and wps is number of writes per second. Signed-off-by: Andrei Gudkov <gudkov.andrei@huawei.com> Reviewed-by: Hyman Huang <yong.huang@smartx.com> Message-Id: <d802e6b8053eb60fbec1a784cf86f67d9528e0a8.1693895970.git.gudkov.andrei@huawei.com> Signed-off-by: Hyman Huang <yong.huang@smartx.com>
2023-09-05 10:05:43 +03:00
DirtyStat.calc_time_ms = dirty_stat_wait(config.calc_time_ms,
initial_time);
rcu_read_lock();
if (!compare_page_hash_info(block_dinfo, block_count)) {
goto out;
}
migration/calc-dirty-rate: millisecond-granularity period This patch allows to measure dirty page rate for sub-second intervals of time. An optional argument is introduced -- calc-time-unit. For example: {"execute": "calc-dirty-rate", "arguments": {"calc-time": 500, "calc-time-unit": "millisecond"} } Millisecond granularity allows to make predictions whether migration will succeed or not. To do this, calculate dirty rate with calc-time set to max allowed downtime (e.g. 300ms), convert measured rate into volume of dirtied memory, and divide by network throughput. If the value is lower than max allowed downtime, then migration will converge. Measurement results for single thread randomly writing to a 1/4/24GiB memory region: +----------------+-----------------------------------------------+ | calc-time | dirty rate MiB/s | | (milliseconds) +----------------+---------------+--------------+ | | theoretical | page-sampling | dirty-bitmap | | | (at 3M wr/sec) | | | +----------------+----------------+---------------+--------------+ | 1GiB | +----------------+----------------+---------------+--------------+ | 100 | 6996 | 7100 | 3192 | | 200 | 4606 | 4660 | 2655 | | 300 | 3305 | 3280 | 2371 | | 400 | 2534 | 2525 | 2154 | | 500 | 2041 | 2044 | 1871 | | 750 | 1365 | 1341 | 1358 | | 1000 | 1024 | 1052 | 1025 | | 1500 | 683 | 678 | 684 | | 2000 | 512 | 507 | 513 | +----------------+----------------+---------------+--------------+ | 4GiB | +----------------+----------------+---------------+--------------+ | 100 | 10232 | 8880 | 4070 | | 200 | 8954 | 8049 | 3195 | | 300 | 7889 | 7193 | 2881 | | 400 | 6996 | 6530 | 2700 | | 500 | 6245 | 5772 | 2312 | | 750 | 4829 | 4586 | 2465 | | 1000 | 3865 | 3780 | 2178 | | 1500 | 2694 | 2633 | 2004 | | 2000 | 2041 | 2031 | 1789 | +----------------+----------------+---------------+--------------+ | 24GiB | +----------------+----------------+---------------+--------------+ | 100 | 11495 | 8640 | 5597 | | 200 | 11226 | 8616 | 3527 | | 300 | 10965 | 8386 | 2355 | | 400 | 10713 | 8370 | 2179 | | 500 | 10469 | 8196 | 2098 | | 750 | 9890 | 7885 | 2556 | | 1000 | 9354 | 7506 | 2084 | | 1500 | 8397 | 6944 | 2075 | | 2000 | 7574 | 6402 | 2062 | +----------------+----------------+---------------+--------------+ Theoretical values are computed according to the following formula: size * (1 - (1-(4096/size))^(time*wps)) / (time * 2^20), where size is in bytes, time is in seconds, and wps is number of writes per second. Signed-off-by: Andrei Gudkov <gudkov.andrei@huawei.com> Reviewed-by: Hyman Huang <yong.huang@smartx.com> Message-Id: <d802e6b8053eb60fbec1a784cf86f67d9528e0a8.1693895970.git.gudkov.andrei@huawei.com> Signed-off-by: Hyman Huang <yong.huang@smartx.com>
2023-09-05 10:05:43 +03:00
update_dirtyrate(DirtyStat.calc_time_ms);
out:
rcu_read_unlock();
free_ramblock_dirty_info(block_dinfo, block_count);
}
static void calculate_dirtyrate(struct DirtyRateConfig config)
{
if (config.mode == DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP) {
calculate_dirtyrate_dirty_bitmap(config);
} else if (config.mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) {
calculate_dirtyrate_dirty_ring(config);
} else {
calculate_dirtyrate_sample_vm(config);
}
trace_dirtyrate_calculate(DirtyStat.dirty_rate);
}
void *get_dirtyrate_thread(void *arg)
{
struct DirtyRateConfig config = *(struct DirtyRateConfig *)arg;
int ret;
rcu_register_thread();
ret = dirtyrate_set_state(&CalculatingState, DIRTY_RATE_STATUS_UNSTARTED,
DIRTY_RATE_STATUS_MEASURING);
if (ret == -1) {
error_report("change dirtyrate state failed.");
return NULL;
}
calculate_dirtyrate(config);
ret = dirtyrate_set_state(&CalculatingState, DIRTY_RATE_STATUS_MEASURING,
DIRTY_RATE_STATUS_MEASURED);
if (ret == -1) {
error_report("change dirtyrate state failed.");
}
rcu_unregister_thread();
return NULL;
}
void qmp_calc_dirty_rate(int64_t calc_time,
migration/calc-dirty-rate: millisecond-granularity period This patch allows to measure dirty page rate for sub-second intervals of time. An optional argument is introduced -- calc-time-unit. For example: {"execute": "calc-dirty-rate", "arguments": {"calc-time": 500, "calc-time-unit": "millisecond"} } Millisecond granularity allows to make predictions whether migration will succeed or not. To do this, calculate dirty rate with calc-time set to max allowed downtime (e.g. 300ms), convert measured rate into volume of dirtied memory, and divide by network throughput. If the value is lower than max allowed downtime, then migration will converge. Measurement results for single thread randomly writing to a 1/4/24GiB memory region: +----------------+-----------------------------------------------+ | calc-time | dirty rate MiB/s | | (milliseconds) +----------------+---------------+--------------+ | | theoretical | page-sampling | dirty-bitmap | | | (at 3M wr/sec) | | | +----------------+----------------+---------------+--------------+ | 1GiB | +----------------+----------------+---------------+--------------+ | 100 | 6996 | 7100 | 3192 | | 200 | 4606 | 4660 | 2655 | | 300 | 3305 | 3280 | 2371 | | 400 | 2534 | 2525 | 2154 | | 500 | 2041 | 2044 | 1871 | | 750 | 1365 | 1341 | 1358 | | 1000 | 1024 | 1052 | 1025 | | 1500 | 683 | 678 | 684 | | 2000 | 512 | 507 | 513 | +----------------+----------------+---------------+--------------+ | 4GiB | +----------------+----------------+---------------+--------------+ | 100 | 10232 | 8880 | 4070 | | 200 | 8954 | 8049 | 3195 | | 300 | 7889 | 7193 | 2881 | | 400 | 6996 | 6530 | 2700 | | 500 | 6245 | 5772 | 2312 | | 750 | 4829 | 4586 | 2465 | | 1000 | 3865 | 3780 | 2178 | | 1500 | 2694 | 2633 | 2004 | | 2000 | 2041 | 2031 | 1789 | +----------------+----------------+---------------+--------------+ | 24GiB | +----------------+----------------+---------------+--------------+ | 100 | 11495 | 8640 | 5597 | | 200 | 11226 | 8616 | 3527 | | 300 | 10965 | 8386 | 2355 | | 400 | 10713 | 8370 | 2179 | | 500 | 10469 | 8196 | 2098 | | 750 | 9890 | 7885 | 2556 | | 1000 | 9354 | 7506 | 2084 | | 1500 | 8397 | 6944 | 2075 | | 2000 | 7574 | 6402 | 2062 | +----------------+----------------+---------------+--------------+ Theoretical values are computed according to the following formula: size * (1 - (1-(4096/size))^(time*wps)) / (time * 2^20), where size is in bytes, time is in seconds, and wps is number of writes per second. Signed-off-by: Andrei Gudkov <gudkov.andrei@huawei.com> Reviewed-by: Hyman Huang <yong.huang@smartx.com> Message-Id: <d802e6b8053eb60fbec1a784cf86f67d9528e0a8.1693895970.git.gudkov.andrei@huawei.com> Signed-off-by: Hyman Huang <yong.huang@smartx.com>
2023-09-05 10:05:43 +03:00
bool has_calc_time_unit,
TimeUnit calc_time_unit,
bool has_sample_pages,
int64_t sample_pages,
bool has_mode,
DirtyRateMeasureMode mode,
Error **errp)
{
static struct DirtyRateConfig config;
QemuThread thread;
int ret;
/*
* If the dirty rate is already being measured, don't attempt to start.
*/
if (qatomic_read(&CalculatingState) == DIRTY_RATE_STATUS_MEASURING) {
error_setg(errp, "the dirty rate is already being measured.");
return;
}
migration/calc-dirty-rate: millisecond-granularity period This patch allows to measure dirty page rate for sub-second intervals of time. An optional argument is introduced -- calc-time-unit. For example: {"execute": "calc-dirty-rate", "arguments": {"calc-time": 500, "calc-time-unit": "millisecond"} } Millisecond granularity allows to make predictions whether migration will succeed or not. To do this, calculate dirty rate with calc-time set to max allowed downtime (e.g. 300ms), convert measured rate into volume of dirtied memory, and divide by network throughput. If the value is lower than max allowed downtime, then migration will converge. Measurement results for single thread randomly writing to a 1/4/24GiB memory region: +----------------+-----------------------------------------------+ | calc-time | dirty rate MiB/s | | (milliseconds) +----------------+---------------+--------------+ | | theoretical | page-sampling | dirty-bitmap | | | (at 3M wr/sec) | | | +----------------+----------------+---------------+--------------+ | 1GiB | +----------------+----------------+---------------+--------------+ | 100 | 6996 | 7100 | 3192 | | 200 | 4606 | 4660 | 2655 | | 300 | 3305 | 3280 | 2371 | | 400 | 2534 | 2525 | 2154 | | 500 | 2041 | 2044 | 1871 | | 750 | 1365 | 1341 | 1358 | | 1000 | 1024 | 1052 | 1025 | | 1500 | 683 | 678 | 684 | | 2000 | 512 | 507 | 513 | +----------------+----------------+---------------+--------------+ | 4GiB | +----------------+----------------+---------------+--------------+ | 100 | 10232 | 8880 | 4070 | | 200 | 8954 | 8049 | 3195 | | 300 | 7889 | 7193 | 2881 | | 400 | 6996 | 6530 | 2700 | | 500 | 6245 | 5772 | 2312 | | 750 | 4829 | 4586 | 2465 | | 1000 | 3865 | 3780 | 2178 | | 1500 | 2694 | 2633 | 2004 | | 2000 | 2041 | 2031 | 1789 | +----------------+----------------+---------------+--------------+ | 24GiB | +----------------+----------------+---------------+--------------+ | 100 | 11495 | 8640 | 5597 | | 200 | 11226 | 8616 | 3527 | | 300 | 10965 | 8386 | 2355 | | 400 | 10713 | 8370 | 2179 | | 500 | 10469 | 8196 | 2098 | | 750 | 9890 | 7885 | 2556 | | 1000 | 9354 | 7506 | 2084 | | 1500 | 8397 | 6944 | 2075 | | 2000 | 7574 | 6402 | 2062 | +----------------+----------------+---------------+--------------+ Theoretical values are computed according to the following formula: size * (1 - (1-(4096/size))^(time*wps)) / (time * 2^20), where size is in bytes, time is in seconds, and wps is number of writes per second. Signed-off-by: Andrei Gudkov <gudkov.andrei@huawei.com> Reviewed-by: Hyman Huang <yong.huang@smartx.com> Message-Id: <d802e6b8053eb60fbec1a784cf86f67d9528e0a8.1693895970.git.gudkov.andrei@huawei.com> Signed-off-by: Hyman Huang <yong.huang@smartx.com>
2023-09-05 10:05:43 +03:00
int64_t calc_time_ms = convert_time_unit(
calc_time,
has_calc_time_unit ? calc_time_unit : TIME_UNIT_SECOND,
TIME_UNIT_MILLISECOND
);
if (!is_calc_time_valid(calc_time_ms)) {
error_setg(errp, "Calculation time is out of range [%dms, %dms].",
MIN_CALC_TIME_MS, MAX_CALC_TIME_MS);
return;
}
if (!has_mode) {
mode = DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING;
}
if (has_sample_pages && mode != DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING) {
error_setg(errp, "sample-pages is used only in page-sampling mode");
return;
}
if (has_sample_pages) {
if (!is_sample_pages_valid(sample_pages)) {
error_setg(errp, "sample-pages is out of range[%d, %d].",
MIN_SAMPLE_PAGE_COUNT,
MAX_SAMPLE_PAGE_COUNT);
return;
}
} else {
sample_pages = DIRTYRATE_DEFAULT_SAMPLE_PAGES;
}
/*
* dirty ring mode only works when kvm dirty ring is enabled.
* on the contrary, dirty bitmap mode is not.
*/
if (((mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) &&
!kvm_dirty_ring_enabled()) ||
((mode == DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP) &&
kvm_dirty_ring_enabled())) {
error_setg(errp, "mode %s is not enabled, use other method instead.",
DirtyRateMeasureMode_str(mode));
return;
}
/*
* Init calculation state as unstarted.
*/
ret = dirtyrate_set_state(&CalculatingState, CalculatingState,
DIRTY_RATE_STATUS_UNSTARTED);
if (ret == -1) {
error_setg(errp, "init dirty rate calculation state failed.");
return;
}
migration/calc-dirty-rate: millisecond-granularity period This patch allows to measure dirty page rate for sub-second intervals of time. An optional argument is introduced -- calc-time-unit. For example: {"execute": "calc-dirty-rate", "arguments": {"calc-time": 500, "calc-time-unit": "millisecond"} } Millisecond granularity allows to make predictions whether migration will succeed or not. To do this, calculate dirty rate with calc-time set to max allowed downtime (e.g. 300ms), convert measured rate into volume of dirtied memory, and divide by network throughput. If the value is lower than max allowed downtime, then migration will converge. Measurement results for single thread randomly writing to a 1/4/24GiB memory region: +----------------+-----------------------------------------------+ | calc-time | dirty rate MiB/s | | (milliseconds) +----------------+---------------+--------------+ | | theoretical | page-sampling | dirty-bitmap | | | (at 3M wr/sec) | | | +----------------+----------------+---------------+--------------+ | 1GiB | +----------------+----------------+---------------+--------------+ | 100 | 6996 | 7100 | 3192 | | 200 | 4606 | 4660 | 2655 | | 300 | 3305 | 3280 | 2371 | | 400 | 2534 | 2525 | 2154 | | 500 | 2041 | 2044 | 1871 | | 750 | 1365 | 1341 | 1358 | | 1000 | 1024 | 1052 | 1025 | | 1500 | 683 | 678 | 684 | | 2000 | 512 | 507 | 513 | +----------------+----------------+---------------+--------------+ | 4GiB | +----------------+----------------+---------------+--------------+ | 100 | 10232 | 8880 | 4070 | | 200 | 8954 | 8049 | 3195 | | 300 | 7889 | 7193 | 2881 | | 400 | 6996 | 6530 | 2700 | | 500 | 6245 | 5772 | 2312 | | 750 | 4829 | 4586 | 2465 | | 1000 | 3865 | 3780 | 2178 | | 1500 | 2694 | 2633 | 2004 | | 2000 | 2041 | 2031 | 1789 | +----------------+----------------+---------------+--------------+ | 24GiB | +----------------+----------------+---------------+--------------+ | 100 | 11495 | 8640 | 5597 | | 200 | 11226 | 8616 | 3527 | | 300 | 10965 | 8386 | 2355 | | 400 | 10713 | 8370 | 2179 | | 500 | 10469 | 8196 | 2098 | | 750 | 9890 | 7885 | 2556 | | 1000 | 9354 | 7506 | 2084 | | 1500 | 8397 | 6944 | 2075 | | 2000 | 7574 | 6402 | 2062 | +----------------+----------------+---------------+--------------+ Theoretical values are computed according to the following formula: size * (1 - (1-(4096/size))^(time*wps)) / (time * 2^20), where size is in bytes, time is in seconds, and wps is number of writes per second. Signed-off-by: Andrei Gudkov <gudkov.andrei@huawei.com> Reviewed-by: Hyman Huang <yong.huang@smartx.com> Message-Id: <d802e6b8053eb60fbec1a784cf86f67d9528e0a8.1693895970.git.gudkov.andrei@huawei.com> Signed-off-by: Hyman Huang <yong.huang@smartx.com>
2023-09-05 10:05:43 +03:00
config.calc_time_ms = calc_time_ms;
config.sample_pages_per_gigabytes = sample_pages;
config.mode = mode;
cleanup_dirtyrate_stat(config);
/*
* update dirty rate mode so that we can figure out what mode has
* been used in last calculation
**/
dirtyrate_mode = mode;
init_dirtyrate_stat(config);
qemu_thread_create(&thread, "get_dirtyrate", get_dirtyrate_thread,
(void *)&config, QEMU_THREAD_DETACHED);
}
migration/calc-dirty-rate: millisecond-granularity period This patch allows to measure dirty page rate for sub-second intervals of time. An optional argument is introduced -- calc-time-unit. For example: {"execute": "calc-dirty-rate", "arguments": {"calc-time": 500, "calc-time-unit": "millisecond"} } Millisecond granularity allows to make predictions whether migration will succeed or not. To do this, calculate dirty rate with calc-time set to max allowed downtime (e.g. 300ms), convert measured rate into volume of dirtied memory, and divide by network throughput. If the value is lower than max allowed downtime, then migration will converge. Measurement results for single thread randomly writing to a 1/4/24GiB memory region: +----------------+-----------------------------------------------+ | calc-time | dirty rate MiB/s | | (milliseconds) +----------------+---------------+--------------+ | | theoretical | page-sampling | dirty-bitmap | | | (at 3M wr/sec) | | | +----------------+----------------+---------------+--------------+ | 1GiB | +----------------+----------------+---------------+--------------+ | 100 | 6996 | 7100 | 3192 | | 200 | 4606 | 4660 | 2655 | | 300 | 3305 | 3280 | 2371 | | 400 | 2534 | 2525 | 2154 | | 500 | 2041 | 2044 | 1871 | | 750 | 1365 | 1341 | 1358 | | 1000 | 1024 | 1052 | 1025 | | 1500 | 683 | 678 | 684 | | 2000 | 512 | 507 | 513 | +----------------+----------------+---------------+--------------+ | 4GiB | +----------------+----------------+---------------+--------------+ | 100 | 10232 | 8880 | 4070 | | 200 | 8954 | 8049 | 3195 | | 300 | 7889 | 7193 | 2881 | | 400 | 6996 | 6530 | 2700 | | 500 | 6245 | 5772 | 2312 | | 750 | 4829 | 4586 | 2465 | | 1000 | 3865 | 3780 | 2178 | | 1500 | 2694 | 2633 | 2004 | | 2000 | 2041 | 2031 | 1789 | +----------------+----------------+---------------+--------------+ | 24GiB | +----------------+----------------+---------------+--------------+ | 100 | 11495 | 8640 | 5597 | | 200 | 11226 | 8616 | 3527 | | 300 | 10965 | 8386 | 2355 | | 400 | 10713 | 8370 | 2179 | | 500 | 10469 | 8196 | 2098 | | 750 | 9890 | 7885 | 2556 | | 1000 | 9354 | 7506 | 2084 | | 1500 | 8397 | 6944 | 2075 | | 2000 | 7574 | 6402 | 2062 | +----------------+----------------+---------------+--------------+ Theoretical values are computed according to the following formula: size * (1 - (1-(4096/size))^(time*wps)) / (time * 2^20), where size is in bytes, time is in seconds, and wps is number of writes per second. Signed-off-by: Andrei Gudkov <gudkov.andrei@huawei.com> Reviewed-by: Hyman Huang <yong.huang@smartx.com> Message-Id: <d802e6b8053eb60fbec1a784cf86f67d9528e0a8.1693895970.git.gudkov.andrei@huawei.com> Signed-off-by: Hyman Huang <yong.huang@smartx.com>
2023-09-05 10:05:43 +03:00
struct DirtyRateInfo *qmp_query_dirty_rate(bool has_calc_time_unit,
TimeUnit calc_time_unit,
Error **errp)
{
migration/calc-dirty-rate: millisecond-granularity period This patch allows to measure dirty page rate for sub-second intervals of time. An optional argument is introduced -- calc-time-unit. For example: {"execute": "calc-dirty-rate", "arguments": {"calc-time": 500, "calc-time-unit": "millisecond"} } Millisecond granularity allows to make predictions whether migration will succeed or not. To do this, calculate dirty rate with calc-time set to max allowed downtime (e.g. 300ms), convert measured rate into volume of dirtied memory, and divide by network throughput. If the value is lower than max allowed downtime, then migration will converge. Measurement results for single thread randomly writing to a 1/4/24GiB memory region: +----------------+-----------------------------------------------+ | calc-time | dirty rate MiB/s | | (milliseconds) +----------------+---------------+--------------+ | | theoretical | page-sampling | dirty-bitmap | | | (at 3M wr/sec) | | | +----------------+----------------+---------------+--------------+ | 1GiB | +----------------+----------------+---------------+--------------+ | 100 | 6996 | 7100 | 3192 | | 200 | 4606 | 4660 | 2655 | | 300 | 3305 | 3280 | 2371 | | 400 | 2534 | 2525 | 2154 | | 500 | 2041 | 2044 | 1871 | | 750 | 1365 | 1341 | 1358 | | 1000 | 1024 | 1052 | 1025 | | 1500 | 683 | 678 | 684 | | 2000 | 512 | 507 | 513 | +----------------+----------------+---------------+--------------+ | 4GiB | +----------------+----------------+---------------+--------------+ | 100 | 10232 | 8880 | 4070 | | 200 | 8954 | 8049 | 3195 | | 300 | 7889 | 7193 | 2881 | | 400 | 6996 | 6530 | 2700 | | 500 | 6245 | 5772 | 2312 | | 750 | 4829 | 4586 | 2465 | | 1000 | 3865 | 3780 | 2178 | | 1500 | 2694 | 2633 | 2004 | | 2000 | 2041 | 2031 | 1789 | +----------------+----------------+---------------+--------------+ | 24GiB | +----------------+----------------+---------------+--------------+ | 100 | 11495 | 8640 | 5597 | | 200 | 11226 | 8616 | 3527 | | 300 | 10965 | 8386 | 2355 | | 400 | 10713 | 8370 | 2179 | | 500 | 10469 | 8196 | 2098 | | 750 | 9890 | 7885 | 2556 | | 1000 | 9354 | 7506 | 2084 | | 1500 | 8397 | 6944 | 2075 | | 2000 | 7574 | 6402 | 2062 | +----------------+----------------+---------------+--------------+ Theoretical values are computed according to the following formula: size * (1 - (1-(4096/size))^(time*wps)) / (time * 2^20), where size is in bytes, time is in seconds, and wps is number of writes per second. Signed-off-by: Andrei Gudkov <gudkov.andrei@huawei.com> Reviewed-by: Hyman Huang <yong.huang@smartx.com> Message-Id: <d802e6b8053eb60fbec1a784cf86f67d9528e0a8.1693895970.git.gudkov.andrei@huawei.com> Signed-off-by: Hyman Huang <yong.huang@smartx.com>
2023-09-05 10:05:43 +03:00
return query_dirty_rate_info(
has_calc_time_unit ? calc_time_unit : TIME_UNIT_SECOND);
}
void hmp_info_dirty_rate(Monitor *mon, const QDict *qdict)
{
migration/calc-dirty-rate: millisecond-granularity period This patch allows to measure dirty page rate for sub-second intervals of time. An optional argument is introduced -- calc-time-unit. For example: {"execute": "calc-dirty-rate", "arguments": {"calc-time": 500, "calc-time-unit": "millisecond"} } Millisecond granularity allows to make predictions whether migration will succeed or not. To do this, calculate dirty rate with calc-time set to max allowed downtime (e.g. 300ms), convert measured rate into volume of dirtied memory, and divide by network throughput. If the value is lower than max allowed downtime, then migration will converge. Measurement results for single thread randomly writing to a 1/4/24GiB memory region: +----------------+-----------------------------------------------+ | calc-time | dirty rate MiB/s | | (milliseconds) +----------------+---------------+--------------+ | | theoretical | page-sampling | dirty-bitmap | | | (at 3M wr/sec) | | | +----------------+----------------+---------------+--------------+ | 1GiB | +----------------+----------------+---------------+--------------+ | 100 | 6996 | 7100 | 3192 | | 200 | 4606 | 4660 | 2655 | | 300 | 3305 | 3280 | 2371 | | 400 | 2534 | 2525 | 2154 | | 500 | 2041 | 2044 | 1871 | | 750 | 1365 | 1341 | 1358 | | 1000 | 1024 | 1052 | 1025 | | 1500 | 683 | 678 | 684 | | 2000 | 512 | 507 | 513 | +----------------+----------------+---------------+--------------+ | 4GiB | +----------------+----------------+---------------+--------------+ | 100 | 10232 | 8880 | 4070 | | 200 | 8954 | 8049 | 3195 | | 300 | 7889 | 7193 | 2881 | | 400 | 6996 | 6530 | 2700 | | 500 | 6245 | 5772 | 2312 | | 750 | 4829 | 4586 | 2465 | | 1000 | 3865 | 3780 | 2178 | | 1500 | 2694 | 2633 | 2004 | | 2000 | 2041 | 2031 | 1789 | +----------------+----------------+---------------+--------------+ | 24GiB | +----------------+----------------+---------------+--------------+ | 100 | 11495 | 8640 | 5597 | | 200 | 11226 | 8616 | 3527 | | 300 | 10965 | 8386 | 2355 | | 400 | 10713 | 8370 | 2179 | | 500 | 10469 | 8196 | 2098 | | 750 | 9890 | 7885 | 2556 | | 1000 | 9354 | 7506 | 2084 | | 1500 | 8397 | 6944 | 2075 | | 2000 | 7574 | 6402 | 2062 | +----------------+----------------+---------------+--------------+ Theoretical values are computed according to the following formula: size * (1 - (1-(4096/size))^(time*wps)) / (time * 2^20), where size is in bytes, time is in seconds, and wps is number of writes per second. Signed-off-by: Andrei Gudkov <gudkov.andrei@huawei.com> Reviewed-by: Hyman Huang <yong.huang@smartx.com> Message-Id: <d802e6b8053eb60fbec1a784cf86f67d9528e0a8.1693895970.git.gudkov.andrei@huawei.com> Signed-off-by: Hyman Huang <yong.huang@smartx.com>
2023-09-05 10:05:43 +03:00
DirtyRateInfo *info = query_dirty_rate_info(TIME_UNIT_SECOND);
monitor_printf(mon, "Status: %s\n",
DirtyRateStatus_str(info->status));
monitor_printf(mon, "Start Time: %"PRIi64" (ms)\n",
info->start_time);
if (info->mode == DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING) {
monitor_printf(mon, "Sample Pages: %"PRIu64" (per GB)\n",
info->sample_pages);
}
monitor_printf(mon, "Period: %"PRIi64" (sec)\n",
info->calc_time);
monitor_printf(mon, "Mode: %s\n",
DirtyRateMeasureMode_str(info->mode));
monitor_printf(mon, "Dirty rate: ");
if (info->has_dirty_rate) {
monitor_printf(mon, "%"PRIi64" (MB/s)\n", info->dirty_rate);
if (info->has_vcpu_dirty_rate) {
DirtyRateVcpuList *rate, *head = info->vcpu_dirty_rate;
for (rate = head; rate != NULL; rate = rate->next) {
monitor_printf(mon, "vcpu[%"PRIi64"], Dirty rate: %"PRIi64
" (MB/s)\n", rate->value->id,
rate->value->dirty_rate);
}
}
} else {
monitor_printf(mon, "(not ready)\n");
}
qapi_free_DirtyRateVcpuList(info->vcpu_dirty_rate);
g_free(info);
}
void hmp_calc_dirty_rate(Monitor *mon, const QDict *qdict)
{
int64_t sec = qdict_get_try_int(qdict, "second", 0);
int64_t sample_pages = qdict_get_try_int(qdict, "sample_pages_per_GB", -1);
bool has_sample_pages = (sample_pages != -1);
bool dirty_ring = qdict_get_try_bool(qdict, "dirty_ring", false);
bool dirty_bitmap = qdict_get_try_bool(qdict, "dirty_bitmap", false);
DirtyRateMeasureMode mode = DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING;
Error *err = NULL;
if (!sec) {
monitor_printf(mon, "Incorrect period length specified!\n");
return;
}
if (dirty_ring && dirty_bitmap) {
monitor_printf(mon, "Either dirty ring or dirty bitmap "
"can be specified!\n");
return;
}
if (dirty_bitmap) {
mode = DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP;
} else if (dirty_ring) {
mode = DIRTY_RATE_MEASURE_MODE_DIRTY_RING;
}
migration/calc-dirty-rate: millisecond-granularity period This patch allows to measure dirty page rate for sub-second intervals of time. An optional argument is introduced -- calc-time-unit. For example: {"execute": "calc-dirty-rate", "arguments": {"calc-time": 500, "calc-time-unit": "millisecond"} } Millisecond granularity allows to make predictions whether migration will succeed or not. To do this, calculate dirty rate with calc-time set to max allowed downtime (e.g. 300ms), convert measured rate into volume of dirtied memory, and divide by network throughput. If the value is lower than max allowed downtime, then migration will converge. Measurement results for single thread randomly writing to a 1/4/24GiB memory region: +----------------+-----------------------------------------------+ | calc-time | dirty rate MiB/s | | (milliseconds) +----------------+---------------+--------------+ | | theoretical | page-sampling | dirty-bitmap | | | (at 3M wr/sec) | | | +----------------+----------------+---------------+--------------+ | 1GiB | +----------------+----------------+---------------+--------------+ | 100 | 6996 | 7100 | 3192 | | 200 | 4606 | 4660 | 2655 | | 300 | 3305 | 3280 | 2371 | | 400 | 2534 | 2525 | 2154 | | 500 | 2041 | 2044 | 1871 | | 750 | 1365 | 1341 | 1358 | | 1000 | 1024 | 1052 | 1025 | | 1500 | 683 | 678 | 684 | | 2000 | 512 | 507 | 513 | +----------------+----------------+---------------+--------------+ | 4GiB | +----------------+----------------+---------------+--------------+ | 100 | 10232 | 8880 | 4070 | | 200 | 8954 | 8049 | 3195 | | 300 | 7889 | 7193 | 2881 | | 400 | 6996 | 6530 | 2700 | | 500 | 6245 | 5772 | 2312 | | 750 | 4829 | 4586 | 2465 | | 1000 | 3865 | 3780 | 2178 | | 1500 | 2694 | 2633 | 2004 | | 2000 | 2041 | 2031 | 1789 | +----------------+----------------+---------------+--------------+ | 24GiB | +----------------+----------------+---------------+--------------+ | 100 | 11495 | 8640 | 5597 | | 200 | 11226 | 8616 | 3527 | | 300 | 10965 | 8386 | 2355 | | 400 | 10713 | 8370 | 2179 | | 500 | 10469 | 8196 | 2098 | | 750 | 9890 | 7885 | 2556 | | 1000 | 9354 | 7506 | 2084 | | 1500 | 8397 | 6944 | 2075 | | 2000 | 7574 | 6402 | 2062 | +----------------+----------------+---------------+--------------+ Theoretical values are computed according to the following formula: size * (1 - (1-(4096/size))^(time*wps)) / (time * 2^20), where size is in bytes, time is in seconds, and wps is number of writes per second. Signed-off-by: Andrei Gudkov <gudkov.andrei@huawei.com> Reviewed-by: Hyman Huang <yong.huang@smartx.com> Message-Id: <d802e6b8053eb60fbec1a784cf86f67d9528e0a8.1693895970.git.gudkov.andrei@huawei.com> Signed-off-by: Hyman Huang <yong.huang@smartx.com>
2023-09-05 10:05:43 +03:00
qmp_calc_dirty_rate(sec, /* calc-time */
false, TIME_UNIT_SECOND, /* calc-time-unit */
has_sample_pages, sample_pages,
true, mode,
&err);
if (err) {
hmp_handle_error(mon, err);
return;
}
monitor_printf(mon, "Starting dirty rate measurement with period %"PRIi64
" seconds\n", sec);
monitor_printf(mon, "[Please use 'info dirty_rate' to check results]\n");
}