Make windowsaio match iometer performance. Without this, we see around 60k on local...

diff --git a/backend.c b/backend.c
index 65a3e184a6d77d7d5439f23c9a4c17813a405326..a0b2c33937ad29afb0c3c7abc04e5d5b9a1cd943 100644 (file)
--- a/backend.c
+++ b/backend.c
@@ -428,7 +428,7 @@ static int wait_for_completions(struct thread_data *td, struct timeval *time)
         * if the queue is full, we MUST reap at least 1 event
         */
        min_evts = min(td->o.iodepth_batch_complete, td->cur_depth);
-       if (full && !min_evts)
+    if ((full && !min_evts) || !td->o.iodepth_batch_complete)
                min_evts = 1;
 
        if (time && (__should_check_rate(td, DDIR_READ) ||

diff --git a/engines/windowsaio.c b/engines/windowsaio.c
index ec8222c90c9f89cba4d16e2f7f5e9f336a639c14..cbbed6a0cf19603022b39ef7179585932d4f0317 100644 (file)
--- a/engines/windowsaio.c
+++ b/engines/windowsaio.c
@@ -284,14 +284,13 @@ static int fio_windowsaio_getevents(struct thread_data *td, unsigned int min,
 
                        if (fov->io_complete) {
                                fov->io_complete = FALSE;
-                               ResetEvent(fov->o.hEvent);
                                wd->aio_events[dequeued] = io_u;
                                dequeued++;
                        }
 
-                       if (dequeued >= min)
-                               break;
                }
+               if (dequeued >= min)
+                       break;
 
                if (dequeued < min) {
                        status = WaitForSingleObject(wd->iocomplete_event, mswait);
@@ -310,23 +309,22 @@ static int fio_windowsaio_queue(struct thread_data *td, struct io_u *io_u)
 {
        struct fio_overlapped *o = io_u->engine_data;
        LPOVERLAPPED lpOvl = &o->o;
-       DWORD iobytes;
        BOOL success = FALSE;
        int rc = FIO_Q_COMPLETED;
 
        fio_ro_check(td, io_u);
 
-       lpOvl->Internal = STATUS_PENDING;
+       lpOvl->Internal = 0;
        lpOvl->InternalHigh = 0;
        lpOvl->Offset = io_u->offset & 0xFFFFFFFF;
        lpOvl->OffsetHigh = io_u->offset >> 32;
 
        switch (io_u->ddir) {
        case DDIR_WRITE:
-               success = WriteFile(io_u->file->hFile, io_u->xfer_buf, io_u->xfer_buflen, &iobytes, lpOvl);
+               success = WriteFile(io_u->file->hFile, io_u->xfer_buf, io_u->xfer_buflen, NULL, lpOvl);
                break;
        case DDIR_READ:
-               success = ReadFile(io_u->file->hFile, io_u->xfer_buf, io_u->xfer_buflen, &iobytes, lpOvl);
+               success = ReadFile(io_u->file->hFile, io_u->xfer_buf, io_u->xfer_buflen, NULL, lpOvl);
                break;
        case DDIR_SYNC:
        case DDIR_DATASYNC:
@@ -403,7 +401,6 @@ static void fio_windowsaio_io_u_free(struct thread_data *td, struct io_u *io_u)
        struct fio_overlapped *o = io_u->engine_data;
 
        if (o) {
-               CloseHandle(o->o.hEvent);
                io_u->engine_data = NULL;
                free(o);
        }
@@ -416,13 +413,7 @@ static int fio_windowsaio_io_u_init(struct thread_data *td, struct io_u *io_u)
        o = malloc(sizeof(*o));
        o->io_complete = FALSE;
        o->io_u = io_u;
-       o->o.hEvent = CreateEvent(NULL, TRUE, FALSE, NULL);
-       if (o->o.hEvent == NULL) {
-               log_err("windowsaio: failed to create event handle\n");
-               free(o);
-               return 1;
-       }
-
+       o->o.hEvent = NULL;
        io_u->engine_data = o;
        return 0;
 }