[-next,v7,0/3] support concurrent sync io for bfq on a specail occasion

Message

Yu Kuai May 28, 2022, 9:50 a.m. UTC

Changes in v7:
 - fix mismatch bfq_inc/del_busy_queues() and bfqq_add/del_bfqq_busy(),
 also retest this patchset on v5.18 to make sure functionality is
 correct.
 - move the updating of 'bfqd->busy_queues' into new apis

Changes in v6:
 - add reviewed-by tag for patch 1

Changes in v5:
 - rename bfq_add_busy_queues() to bfq_inc_busy_queues() in patch 1
 - fix wrong definition in patch 1
 - fix spelling mistake in patch 2: leaset -> least
 - update comments in patch 3
 - add reviewed-by tag in patch 2,3

Changes in v4:
 - split bfq_update_busy_queues() to bfq_add/dec_busy_queues(),
   suggested by Jan Kara.
 - remove unused 'in_groups_with_pending_reqs',

Changes in v3:
 - remove the cleanup patch that is irrelevant now(I'll post it
   separately).
 - instead of hacking wr queues and using weights tree insertion/removal,
   using bfq_add/del_bfqq_busy() to count the number of groups
   (suggested by Jan Kara).

Changes in v2:
 - Use a different approch to count root group, which is much simple.

Currently, bfq can't handle sync io concurrently as long as they
are not issued from root group. This is because
'bfqd->num_groups_with_pending_reqs > 0' is always true in
bfq_asymmetric_scenario().

The way that bfqg is counted into 'num_groups_with_pending_reqs':

Before this patchset:
 1) root group will never be counted.
 2) Count if bfqg or it's child bfqgs have pending requests.
 3) Don't count if bfqg and it's child bfqgs complete all the requests.

After this patchset:
 1) root group is counted.
 2) Count if bfqg have at least one bfqq that is marked busy.
 3) Don't count if bfqg doesn't have any busy bfqqs.

The main reason to use busy state of bfqq instead of 'pending requests'
is that bfqq can stay busy after dispatching the last request if idling
is needed for service guarantees.

With the above changes, concurrent sync io can be supported if only
one group is activated.

fio test script(startdelay is used to avoid queue merging):
[global]
filename=/dev/sda
allow_mounted_write=0
ioengine=psync
direct=1
ioscheduler=bfq
offset_increment=10g
group_reporting
rw=randwrite
bs=4k

[test1]
numjobs=1

[test2]
startdelay=1
numjobs=1

[test3]
startdelay=2
numjobs=1

[test4]
startdelay=3
numjobs=1

[test5]
startdelay=4
numjobs=1

[test6]
startdelay=5
numjobs=1

[test7]
startdelay=6
numjobs=1

[test8]
startdelay=7
numjobs=1

test result:
running fio on root cgroup
v5.18:	   112 Mib/s
v5.18-patched: 112 Mib/s

running fio on non-root cgroup
v5.18:	   51.2 Mib/s
v5.18-patched: 112 Mib/s

Note that I also test null_blk with "irqmode=2
completion_nsec=100000000(100ms) hw_queue_depth=1", and tests show
that service guarantees are still preserved.

Previous versions:
RFC: https://lore.kernel.org/all/[e-mail]/
v1: https://lore.kernel.org/all/[e-mail]/
v2: https://lore.kernel.org/all/[e-mail]/
v3: https://lore.kernel.org/all/[e-mail]/
v4: https://lore.kernel.org/all/[e-mail]/
v5: https://lore.kernel.org/all/[e-mail]/
v6: https://lore.kernel.org/all/[e-mail]/

Yu Kuai (3):
  block, bfq: record how many queues are busy in bfq_group
  block, bfq: refactor the counting of 'num_groups_with_pending_reqs'
  block, bfq: do not idle if only one group is activated

 block/bfq-cgroup.c  |  1 +
 block/bfq-iosched.c | 48 +++-----------------------------------
 block/bfq-iosched.h | 57 +++++++--------------------------------------
 block/bfq-wf2q.c    | 41 ++++++++++++++++++++------------
 4 files changed, 39 insertions(+), 108 deletions(-)