releases/4.19.51/block-bfq-increase-idling-for-weight-raised-queues.patch

   1 From b96ccad0ca65c47bfc56df7de2f8fbdf90fff3e7 Mon Sep 17 00:00:00 2001
   2 From: Paolo Valente <paolo.valente@linaro.org>
   3 Date: Tue, 12 Mar 2019 09:59:27 +0100
   4 Subject: block, bfq: increase idling for weight-raised queues
   5 MIME-Version: 1.0
   6 Content-Type: text/plain; charset=UTF-8
   7 Content-Transfer-Encoding: 8bit
   8
   9 [ Upstream commit 778c02a236a8728bb992de10ed1f12c0be5b7b0e ]
  10
  11 If a sync bfq_queue has a higher weight than some other queue, and
  12 remains temporarily empty while in service, then, to preserve the
  13 bandwidth share of the queue, it is necessary to plug I/O dispatching
  14 until a new request arrives for the queue. In addition, a timeout
  15 needs to be set, to avoid waiting for ever if the process associated
  16 with the queue has actually finished its I/O.
  17
  18 Even with the above timeout, the device is however not fed with new
  19 I/O for a while, if the process has finished its I/O. If this happens
  20 often, then throughput drops and latencies grow. For this reason, the
  21 timeout is kept rather low: 8 ms is the current default.
  22
  23 Unfortunately, such a low value may cause, on the opposite end, a
  24 violation of bandwidth guarantees for a process that happens to issue
  25 new I/O too late. The higher the system load, the higher the
  26 probability that this happens to some process. This is a problem in
  27 scenarios where service guarantees matter more than throughput. One
  28 important case are weight-raised queues, which need to be granted a
  29 very high fraction of the bandwidth.
  30
  31 To address this issue, this commit lower-bounds the plugging timeout
  32 for weight-raised queues to 20 ms. This simple change provides
  33 relevant benefits. For example, on a PLEXTOR PX-256M5S, with which
  34 gnome-terminal starts in 0.6 seconds if there is no other I/O in
  35 progress, the same applications starts in
  36 - 0.8 seconds, instead of 1.2 seconds, if ten files are being read
  37   sequentially in parallel
  38 - 1 second, instead of 2 seconds, if, in parallel, five files are
  39   being read sequentially, and five more files are being written
  40   sequentially
  41
  42 Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
  43 Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name>
  44 Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
  45 Signed-off-by: Jens Axboe <axboe@kernel.dk>
  46 Signed-off-by: Sasha Levin <sashal@kernel.org>
  47 ---
  48  block/bfq-iosched.c | 2 ++
  49  1 file changed, 2 insertions(+)
  50
  51 diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c
  52 index 15e8c9955b79..6bb397995610 100644
  53 --- a/block/bfq-iosched.c
  54 +++ b/block/bfq-iosched.c
  55 @@ -2509,6 +2509,8 @@ static void bfq_arm_slice_timer(struct bfq_data *bfqd)
  56         if (BFQQ_SEEKY(bfqq) && bfqq->wr_coeff == 1 &&
  57             bfq_symmetric_scenario(bfqd))
  58                 sl = min_t(u64, sl, BFQ_MIN_TT);
  59 +       else if (bfqq->wr_coeff > 1)
  60 +               sl = max_t(u32, sl, 20ULL * NSEC_PER_MSEC);
  61
  62         bfqd->last_idling_start = ktime_get();
  63         hrtimer_start(&bfqd->idle_slice_timer, ns_to_ktime(sl),
  64 --
  65 2.20.1
  66