Re: [PATCH v6 3/4] block: add block timer and block throttling algorithm

[Stefan Hajnoczi <stefanha@gmail.com>]

On Thu, Sep 1, 2011 at 12:44 PM, Zhi Yong Wu <wuzhy@linux.vnet.ibm.com> wrote:
> Note:
>     1.) When bps/iops limits are specified to a small value such as 511 bytes/s, this VM will hang up. We are considering how to handle this senario.
>     2.) When "dd" command is issued in guest, if its option bs is set to a large value such as "bs=1024K", the result speed will slightly bigger than the limits.
>
> For these problems, if you have nice thought, pls let us know.:)
>
> Signed-off-by: Zhi Yong Wu <wuzhy@linux.vnet.ibm.com>
> ---
>  block.c     |  290 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++--
>  block.h     |    5 +
>  block_int.h |    9 ++
>  3 files changed, 296 insertions(+), 8 deletions(-)
>
> diff --git a/block.c b/block.c
> index 17ee3df..680f1e7 100644
> --- a/block.c
> +++ b/block.c
> @@ -30,6 +30,9 @@
>  #include "qemu-objects.h"
>  #include "qemu-coroutine.h"
>
> +#include "qemu-timer.h"
> +#include "block/blk-queue.h"
> +
>  #ifdef CONFIG_BSD
>  #include <sys/types.h>
>  #include <sys/stat.h>
> @@ -72,6 +75,13 @@ static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
>                                          QEMUIOVector *iov);
>  static int coroutine_fn bdrv_co_flush_em(BlockDriverState *bs);
>
> +static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
> +        bool is_write, double elapsed_time, uint64_t *wait);
> +static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
> +        double elapsed_time, uint64_t *wait);
> +static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
> +        bool is_write, uint64_t *wait);
> +
>  static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
>     QTAILQ_HEAD_INITIALIZER(bdrv_states);
>
> @@ -104,6 +114,64 @@ int is_windows_drive(const char *filename)
>  }
>  #endif
>
> +/* throttling disk I/O limits */
> +void bdrv_io_limits_disable(BlockDriverState *bs)
> +{
> +    bs->io_limits_enabled = false;
> +
> +    if (bs->block_queue) {
> +        qemu_block_queue_flush(bs->block_queue);
> +        qemu_del_block_queue(bs->block_queue);
> +        bs->block_queue = NULL;
> +    }
> +
> +    if (bs->block_timer) {
> +        qemu_del_timer(bs->block_timer);
> +        qemu_free_timer(bs->block_timer);
> +        bs->block_timer = NULL;
> +    }
> +
> +    bs->slice_start[BLOCK_IO_LIMIT_READ]  = 0;
> +    bs->slice_start[BLOCK_IO_LIMIT_WRITE] = 0;
> +
> +    bs->slice_end[BLOCK_IO_LIMIT_READ]    = 0;
> +    bs->slice_end[BLOCK_IO_LIMIT_WRITE]   = 0;
> +}
> +
> +static void bdrv_block_timer(void *opaque)
> +{
> +    BlockDriverState *bs = opaque;
> +    BlockQueue *queue = bs->block_queue;
> +
> +    qemu_block_queue_flush(queue);
> +}
> +
> +void bdrv_io_limits_enable(BlockDriverState *bs)
> +{
> +    bs->block_queue    = qemu_new_block_queue();
> +    bs->block_timer    = qemu_new_timer_ns(vm_clock, bdrv_block_timer, bs);
> +
> +    bs->slice_start[BLOCK_IO_LIMIT_READ]  = qemu_get_clock_ns(vm_clock);
> +    bs->slice_start[BLOCK_IO_LIMIT_WRITE] =
> +           bs->slice_start[BLOCK_IO_LIMIT_READ];
> +
> +    bs->slice_end[BLOCK_IO_LIMIT_READ]    =
> +           bs->slice_start[BLOCK_IO_LIMIT_READ] + BLOCK_IO_SLICE_TIME;
> +    bs->slice_end[BLOCK_IO_LIMIT_WRITE]   =
> +           bs->slice_end[BLOCK_IO_LIMIT_READ];
> +}
> +
> +bool bdrv_io_limits_enabled(BlockDriverState *bs)
> +{
> +    BlockIOLimit *io_limits = &bs->io_limits;
> +    return io_limits->bps[BLOCK_IO_LIMIT_READ]
> +         || io_limits->bps[BLOCK_IO_LIMIT_WRITE]
> +         || io_limits->bps[BLOCK_IO_LIMIT_TOTAL]
> +         || io_limits->iops[BLOCK_IO_LIMIT_READ]
> +         || io_limits->iops[BLOCK_IO_LIMIT_WRITE]
> +         || io_limits->iops[BLOCK_IO_LIMIT_TOTAL];
> +}
> +
>  /* check if the path starts with "<protocol>:" */
>  static int path_has_protocol(const char *path)
>  {
> @@ -694,6 +762,11 @@ int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
>             bs->change_cb(bs->change_opaque, CHANGE_MEDIA);
>     }
>
> +    /* throttling disk I/O limits */
> +    if (bs->io_limits_enabled) {
> +        bdrv_io_limits_enable(bs);
> +    }
> +
>     return 0;
>
>  unlink_and_fail:
> @@ -732,6 +805,18 @@ void bdrv_close(BlockDriverState *bs)
>         if (bs->change_cb)
>             bs->change_cb(bs->change_opaque, CHANGE_MEDIA);
>     }
> +
> +    /* throttling disk I/O limits */
> +    if (bs->block_queue) {
> +        qemu_del_block_queue(bs->block_queue);
> +        bs->block_queue = NULL;
> +    }
> +
> +    if (bs->block_timer) {
> +        qemu_del_timer(bs->block_timer);
> +        qemu_free_timer(bs->block_timer);
> +        bs->block_timer = NULL;
> +    }
>  }
>
>  void bdrv_close_all(void)
> @@ -2290,13 +2375,29 @@ BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
>                                  BlockDriverCompletionFunc *cb, void *opaque)
>  {
>     BlockDriver *drv = bs->drv;
> +    uint64_t wait_time = 0;
> +    BlockDriverAIOCB *ret;
>
>     trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
>
> -    if (!drv)
> -        return NULL;
> -    if (bdrv_check_request(bs, sector_num, nb_sectors))
> +    if (!drv || bdrv_check_request(bs, sector_num, nb_sectors)) {
>         return NULL;
> +    }
> +
> +    /* throttling disk read I/O */
> +    if (bs->io_limits_enabled) {
> +        if (bdrv_exceed_io_limits(bs, nb_sectors, false, &wait_time)) {
> +            ret = qemu_block_queue_enqueue(bs->block_queue, bs, bdrv_aio_readv,
> +                           sector_num, qiov, nb_sectors, cb, opaque);
> +            qemu_mod_timer(bs->block_timer,
> +                           wait_time + qemu_get_clock_ns(vm_clock));
> +            return ret;
> +        }
> +
> +        bs->io_disps.bytes[BLOCK_IO_LIMIT_READ] +=
> +                           (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
> +        bs->io_disps.ios[BLOCK_IO_LIMIT_READ]++;
> +    }
>
>     return drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
>                                cb, opaque);
> @@ -2345,15 +2446,14 @@ BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
>     BlockDriver *drv = bs->drv;
>     BlockDriverAIOCB *ret;
>     BlockCompleteData *blk_cb_data;
> +    uint64_t wait_time = 0;
>
>     trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
>
> -    if (!drv)
> -        return NULL;
> -    if (bs->read_only)
> -        return NULL;
> -    if (bdrv_check_request(bs, sector_num, nb_sectors))
> +    if (!drv || bs->read_only
> +        || bdrv_check_request(bs, sector_num, nb_sectors)) {
>         return NULL;
> +    }
>
>     if (bs->dirty_bitmap) {
>         blk_cb_data = blk_dirty_cb_alloc(bs, sector_num, nb_sectors, cb,
> @@ -2362,6 +2462,17 @@ BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
>         opaque = blk_cb_data;
>     }
>
> +    /* throttling disk write I/O */
> +    if (bs->io_limits_enabled) {
> +        if (bdrv_exceed_io_limits(bs, nb_sectors, true, &wait_time)) {
> +            ret = qemu_block_queue_enqueue(bs->block_queue, bs, bdrv_aio_writev,
> +                                  sector_num, qiov, nb_sectors, cb, opaque);
> +            qemu_mod_timer(bs->block_timer,
> +                                  wait_time + qemu_get_clock_ns(vm_clock));
> +            return ret;
> +        }
> +    }
> +
>     ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
>                                cb, opaque);
>
> @@ -2369,6 +2480,12 @@ BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
>         if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
>             bs->wr_highest_sector = sector_num + nb_sectors - 1;
>         }
> +
> +        if (bs->io_limits_enabled) {
> +            bs->io_disps.bytes[BLOCK_IO_LIMIT_WRITE] +=
> +                               (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
> +            bs->io_disps.ios[BLOCK_IO_LIMIT_WRITE]++;
> +        }
>     }
>
>     return ret;
> @@ -2633,6 +2750,163 @@ void bdrv_aio_cancel(BlockDriverAIOCB *acb)
>     acb->pool->cancel(acb);
>  }
>
> +static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
> +                 bool is_write, double elapsed_time, uint64_t *wait) {
> +    uint64_t bps_limit = 0;
> +    double   bytes_limit, bytes_disp, bytes_res;
> +    double   slice_time, wait_time;
> +
> +    if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
> +        bps_limit = bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
> +    } else if (bs->io_limits.bps[is_write]) {
> +        bps_limit = bs->io_limits.bps[is_write];
> +    } else {
> +        if (wait) {
> +            *wait = 0;
> +        }
> +
> +        return false;
> +    }
> +
> +    slice_time = bs->slice_end[is_write] - bs->slice_start[is_write];
> +    slice_time /= (NANOSECONDS_PER_SECOND);
> +    bytes_limit = bps_limit * slice_time;
> +    bytes_disp  = bs->io_disps.bytes[is_write];
> +    if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
> +        bytes_disp += bs->io_disps.bytes[!is_write];
> +    }
> +
> +    bytes_res   = (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
> +
> +    if (bytes_disp + bytes_res <= bytes_limit) {
> +        if (wait) {
> +            *wait = 0;
> +        }
> +
> +        return false;
> +    }
> +
> +    /* Calc approx time to dispatch */
> +    wait_time = (bytes_disp + bytes_res) / bps_limit - elapsed_time;
> +
> +    if (wait) {
> +        *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
> +    }
> +
> +    return true;
> +}
> +
> +static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
> +                             double elapsed_time, uint64_t *wait) {
> +    uint64_t iops_limit = 0;
> +    double   ios_limit, ios_disp;
> +    double   slice_time, wait_time;
> +
> +    if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
> +        iops_limit = bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
> +    } else if (bs->io_limits.iops[is_write]) {
> +        iops_limit = bs->io_limits.iops[is_write];
> +    } else {
> +        if (wait) {
> +            *wait = 0;
> +        }
> +
> +        return false;
> +    }
> +
> +    slice_time = bs->slice_end[is_write] - bs->slice_start[is_write];
> +    slice_time /= (NANOSECONDS_PER_SECOND);
> +    ios_limit  = iops_limit * slice_time;
> +    ios_disp   = bs->io_disps.ios[is_write];
> +    if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
> +        ios_disp += bs->io_disps.ios[!is_write];
> +    }
> +
> +    if (ios_disp + 1 <= ios_limit) {
> +        if (wait) {
> +            *wait = 0;
> +        }
> +
> +        return false;
> +    }
> +
> +    /* Calc approx time to dispatch */
> +    wait_time = (ios_disp + 1) / iops_limit;
> +    if (wait_time > elapsed_time) {
> +        wait_time = wait_time - elapsed_time;
> +    } else {
> +        wait_time = 0;
> +    }
> +
> +    if (wait) {
> +        *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
> +    }
> +
> +    return true;
> +}
> +
> +static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
> +                           bool is_write, uint64_t *wait) {
> +    int64_t  now;
> +    uint64_t bps_wait = 0, iops_wait = 0, max_wait;
> +    double   elapsed_time;
> +    int      bps_ret, iops_ret;
> +
> +    now = qemu_get_clock_ns(vm_clock);
> +    if ((bs->slice_start[is_write] < now)
> +        && (bs->slice_end[is_write] > now)) {
> +        bs->slice_end[is_write]   = now + BLOCK_IO_SLICE_TIME;
> +    } else {
> +        bs->slice_start[is_write]     = now;
> +        bs->slice_end[is_write]       = now + BLOCK_IO_SLICE_TIME;
> +
> +        bs->io_disps.bytes[is_write]  = 0;
> +        bs->io_disps.bytes[!is_write] = 0;
> +
> +        bs->io_disps.ios[is_write]    = 0;
> +        bs->io_disps.ios[!is_write]   = 0;

Does it make sense to keep separate slice_start/slice_end for read and
write since we reset the dispatched statistics to zero for both?
Perhaps we should use a scalar slice_start/slice_end and not two
separate values for read/write.

> +    }
> +
> +    /* If a limit was exceeded, immediately queue this request */
> +    if (qemu_block_queue_has_pending(bs->block_queue)) {
> +        if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]
> +            || bs->io_limits.bps[is_write] || bs->io_limits.iops[is_write]
> +            || bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
> +            if (wait) {
> +                *wait = 0;

This causes the queue to be flushed each time the guest enqueues an
I/O while there are queued requests.  Perhaps this is (part of) the
CPU overhead that Ryan's benchmarking discovered.

If we try to preserve request ordering then I don't think there is a
reason to modify the timer once it has been set.

Stefan