diff --git a/Makefile b/Makefile
index 6e822350..87330e74 100644
--- a/Makefile
+++ b/Makefile
@@ -2,7 +2,7 @@ BLOCKSTORE_OBJS := allocator.o blockstore.o blockstore_impl.o blockstore_init.o
 	blockstore_write.o blockstore_sync.o blockstore_stable.o blockstore_rollback.o blockstore_flush.o crc32c.o ringloop.o timerfd_interval.o
 # -fsanitize=address
 CXXFLAGS := -g -O3 -Wall -Wno-sign-compare -Wno-comment -Wno-parentheses -Wno-pointer-arith -fPIC -fdiagnostics-color=always
-all: $(BLOCKSTORE_OBJS) libfio_blockstore.so osd libfio_sec_osd.so test_blockstore stub_osd osd_test
+all: $(BLOCKSTORE_OBJS) libfio_blockstore.so osd libfio_sec_osd.so test_blockstore stub_osd stub_bench osd_test
 clean:
 	rm -f *.o
 
@@ -48,6 +48,8 @@ osd: ./libblockstore.so osd_main.cpp osd.h osd_ops.h $(OSD_OBJS)
 	g++ $(CXXFLAGS) -o osd osd_main.cpp $(OSD_OBJS) ./libblockstore.so -ltcmalloc_minimal -luring
 stub_osd: stub_osd.cpp osd_ops.h rw_blocking.o
 	g++ $(CXXFLAGS) -o stub_osd stub_osd.cpp rw_blocking.o -ltcmalloc_minimal
+stub_bench: stub_bench.cpp osd_ops.h rw_blocking.o
+	g++ $(CXXFLAGS) -o stub_bench stub_bench.cpp rw_blocking.o -ltcmalloc_minimal
 rw_blocking.o: rw_blocking.cpp rw_blocking.h
 	g++ $(CXXFLAGS) -c -o $@ $<
 osd_test: osd_test.cpp osd_ops.h rw_blocking.o
diff --git a/stub_bench.cpp b/stub_bench.cpp
new file mode 100644
index 00000000..2de2113e
--- /dev/null
+++ b/stub_bench.cpp
@@ -0,0 +1,148 @@
+/**
+ * Stub benchmarker
+ */
+
+#include <sys/types.h>
+#include <time.h>
+#include <sys/socket.h>
+#include <netinet/in.h>
+#include <netinet/tcp.h>
+#include <arpa/inet.h>
+#include <string.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <errno.h>
+#include <stdlib.h>
+#include <signal.h>
+
+#include <stdexcept>
+
+#include "rw_blocking.h"
+#include "osd_ops.h"
+
+int connect_stub(const char *server_address, int server_port);
+
+void run_bench(int peer_fd);
+
+static uint64_t write_sum = 0, write_count = 0;
+static uint64_t sync_sum = 0, sync_count = 0;
+
+void handle_sigint(int sig)
+{
+    printf("4k randwrite: %lu us avg\n", write_sum/write_count);
+    printf("sync: %lu us avg\n", sync_sum/sync_count);
+    exit(0);
+}
+
+int main(int narg, char *args[])
+{
+    signal(SIGINT, handle_sigint);
+    int peer_fd = connect_stub("127.0.0.1", 11203);
+    run_bench(peer_fd);
+    close(peer_fd);
+    return 0;
+}
+
+int connect_stub(const char *server_address, int server_port)
+{
+    struct sockaddr_in addr;
+    int r;
+    if ((r = inet_pton(AF_INET, server_address, &addr.sin_addr)) != 1)
+    {
+        fprintf(stderr, "server address: %s%s\n", server_address, r == 0 ? " is not valid" : ": no ipv4 support");
+        return -1;
+    }
+    addr.sin_family = AF_INET;
+    addr.sin_port = htons(server_port);
+    int connect_fd = socket(AF_INET, SOCK_STREAM, 0);
+    if (connect_fd < 0)
+    {
+        perror("socket");
+        return -1;
+    }
+    if (connect(connect_fd, (sockaddr*)&addr, sizeof(addr)) < 0)
+    {
+        perror("connect");
+        return -1;
+    }
+    int one = 1;
+    setsockopt(connect_fd, SOL_TCP, TCP_NODELAY, &one, sizeof(one));
+    return connect_fd;
+}
+
+bool check_reply(int r, osd_any_op_t & op, osd_any_reply_t & reply, int expected)
+{
+    if (r != OSD_PACKET_SIZE)
+    {
+        printf("read failed\n");
+        return false;
+    }
+    if (reply.hdr.magic != SECONDARY_OSD_REPLY_MAGIC ||
+        reply.hdr.id != op.hdr.id || reply.hdr.opcode != op.hdr.opcode)
+    {
+        printf("bad reply: magic, id or opcode does not match request\n");
+        return false;
+    }
+    if (reply.hdr.retval != expected)
+    {
+        printf("operation failed, retval=%ld (%s)\n", reply.hdr.retval, strerror(-reply.hdr.retval));
+        return false;
+    }
+    return true;
+}
+
+void run_bench(int peer_fd)
+{
+    osd_any_op_t op;
+    osd_any_reply_t reply;
+    void *buf = NULL;
+    int r;
+    timespec tv_begin, tv_end;
+    clock_gettime(CLOCK_REALTIME, &tv_begin);
+    while (1)
+    {
+        // write
+        op.hdr.magic = SECONDARY_OSD_OP_MAGIC;
+        op.hdr.id = 1;
+        op.hdr.opcode = OSD_OP_SECONDARY_WRITE;
+        op.sec_rw.oid.inode = 3;
+        op.sec_rw.oid.stripe = (rand() << 17) % (1 << 29); // 512 MB
+        op.sec_rw.version = 0;
+        op.sec_rw.len = 4096;
+        op.sec_rw.offset = (rand() * op.sec_rw.len) % (1 << 17);
+        buf = malloc(op.sec_rw.len);
+        memset(buf, rand() % 255, op.sec_rw.len);
+        r = write_blocking(peer_fd, op.buf, OSD_PACKET_SIZE) == OSD_PACKET_SIZE;
+        if (r)
+            r = write_blocking(peer_fd, buf, op.sec_rw.len) == op.sec_rw.len;
+        free(buf);
+        if (!r)
+            break;
+        r = read_blocking(peer_fd, reply.buf, OSD_PACKET_SIZE);
+        if (!check_reply(r, op, reply, op.sec_rw.len))
+            break;
+        clock_gettime(CLOCK_REALTIME, &tv_end);
+        write_count++;
+        write_sum += (
+            (tv_end.tv_sec - tv_begin.tv_sec)*1000000 +
+            tv_end.tv_nsec/1000 - tv_begin.tv_nsec/1000
+        );
+        // sync/stab
+        op.hdr.magic = SECONDARY_OSD_OP_MAGIC;
+        op.hdr.id = 1;
+        op.hdr.opcode = OSD_OP_TEST_SYNC_STAB_ALL;
+        r = write_blocking(peer_fd, op.buf, OSD_PACKET_SIZE) == OSD_PACKET_SIZE;
+        if (!r)
+            break;
+        r = read_blocking(peer_fd, reply.buf, OSD_PACKET_SIZE);
+        if (!check_reply(r, op, reply, 0))
+            break;
+        clock_gettime(CLOCK_REALTIME, &tv_begin);
+        sync_count++;
+        sync_sum += (
+            (tv_begin.tv_sec - tv_end.tv_sec)*1000000 +
+            tv_begin.tv_nsec/1000 - tv_end.tv_nsec/1000
+        );
+    }
+}