v4: * Added poll time self-tuning algorithm [Christian and Paolo] * Try a single iteration of polling to avoid non-blocking ppoll(2)/epoll_wait(2) [Paolo] * Reordered patches to make performance analysis easier - see below
v3: * Avoid ppoll(2)/epoll_wait(2) if polling succeeded [Paolo] * Disable guest->host virtqueue notification during polling [Christian] * Rebased on top of my virtio-blk/scsi virtqueue notification disable patches v2: * Uninitialized node->deleted gone [Fam] * Removed 1024 polling loop iteration qemu_clock_get_ns() optimization which created a weird step pattern [Fam] * Unified with AioHandler, dropped AioPollHandler struct [Paolo] (actually I think Paolo had more in mind but this is the first step) * Only poll when all event loop resources support it [Paolo] * Added run_poll_handlers_begin/end trace events for perf analysis * Sorry, Christian, no virtqueue kick suppression yet Recent performance investigation work done by Karl Rister shows that the guest->host notification takes around 20 us. This is more than the "overhead" of QEMU itself (e.g. block layer). One way to avoid the costly exit is to use polling instead of notification. The main drawback of polling is that it consumes CPU resources. In order to benefit performance the host must have extra CPU cycles available on physical CPUs that aren't used by the guest. This is an experimental AioContext polling implementation. It adds a polling callback into the event loop. Polling functions are implemented for virtio-blk virtqueue guest->host kick and Linux AIO completion. The -object iothread,poll-max-ns=NUM parameter sets the number of nanoseconds to poll before entering the usual blocking poll(2) syscall. Try setting this parameter to the time from old request completion to new virtqueue kick. By default no polling is done so you must set this parameter to get busy polling. Patch series overview: Here are convenient points in the patch series at which to do performance benchmarking to see how various pieces interact. 1. Basic -iothread poll-max-ns=NUM parameter without self-tuning: aio: add flag to skip fds to aio_dispatch() aio: add AioPollFn and io_poll() interface aio: add polling mode to AioContext virtio: poll virtqueues for new buffers linux-aio: poll ring for completions iothread: add polling parameters 2. My "[PATCH 0/3] virtio: disable notifications in blk and scsi" series is needed as a base. This is unrelated to polling but we'll also disable notifications during polling later: virtio-blk: suppress virtqueue kick during processing virtio-scsi: suppress virtqueue kick during processing 3. Disable virtqueue notifications (reduce vmexits) during polling: virtio: turn vq->notification into a nested counter aio: add .io_poll_begin/end() callbacks virtio: disable virtqueue notifications during polling 4. Self-tuning, adjusts polling time depending on wait times: aio: self-tune polling time iothread: add poll-grow and poll-shrink parameters Stefan Hajnoczi (13): aio: add flag to skip fds to aio_dispatch() aio: add AioPollFn and io_poll() interface aio: add polling mode to AioContext virtio: poll virtqueues for new buffers linux-aio: poll ring for completions iothread: add polling parameters virtio-blk: suppress virtqueue kick during processing virtio-scsi: suppress virtqueue kick during processing virtio: turn vq->notification into a nested counter aio: add .io_poll_begin/end() callbacks virtio: disable virtqueue notifications during polling aio: self-tune polling time iothread: add poll-grow and poll-shrink parameters include/block/aio.h | 53 +++++++- include/sysemu/iothread.h | 5 + aio-posix.c | 308 +++++++++++++++++++++++++++++++++++++++----- aio-win32.c | 32 ++++- async.c | 21 ++- block/curl.c | 8 +- block/iscsi.c | 3 +- block/linux-aio.c | 19 ++- block/nbd-client.c | 8 +- block/nfs.c | 7 +- block/sheepdog.c | 26 ++-- block/ssh.c | 4 +- block/win32-aio.c | 4 +- hw/block/virtio-blk.c | 18 ++- hw/scsi/virtio-scsi.c | 36 +++--- hw/virtio/virtio.c | 54 ++++++-- iohandler.c | 2 +- iothread.c | 84 ++++++++++++ nbd/server.c | 9 +- stubs/set-fd-handler.c | 1 + tests/test-aio.c | 4 +- util/event_notifier-posix.c | 2 +- trace-events | 6 + 23 files changed, 604 insertions(+), 110 deletions(-) -- 2.9.3
