| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <x49y6eylfd4.fsf@segfault.boston.devel.redhat.com>
Date: Tue, 01 Jun 2010 15:31:03 -0400
From: Jeff Moyer <jmoyer@...hat.com>
To: Vivek Goyal <vgoyal@...hat.com>
Cc: linux-kernel@...r.kernel.org, linux-ext4@...r.kernel.org,
axboe@...nel.dk
Subject: Re: [PATCH 1/4] cfq-iosched: Keep track of average think time for the sync-noidle workload.
Vivek Goyal <vgoyal@...hat.com> writes:
> On Tue, May 18, 2010 at 02:20:17PM -0400, Jeff Moyer wrote:
>> This patch uses an average think time for the entirety of the sync-noidle
>> workload to determine whether or not to idle on said workload. This brings
>> it more in line with the policy for the sync queues in the sync workload.
>>
>> Testing shows that this provided an overall increase in throughput for
>> a mixed workload on my hardware RAID array.
>>
>> Signed-off-by: Jeff Moyer <jmoyer@...hat.com>
>> ---
>> block/cfq-iosched.c | 44 +++++++++++++++++++++++++++++++++++++++-----
>> 1 files changed, 39 insertions(+), 5 deletions(-)
>>
>> diff --git a/block/cfq-iosched.c b/block/cfq-iosched.c
>> index 838834b..46a7fe5 100644
>> --- a/block/cfq-iosched.c
>> +++ b/block/cfq-iosched.c
>> @@ -83,9 +83,14 @@ struct cfq_rb_root {
>> unsigned total_weight;
>> u64 min_vdisktime;
>> struct rb_node *active;
>> + unsigned long last_end_request;
>> + unsigned long ttime_total;
>> + unsigned long ttime_samples;
>> + unsigned long ttime_mean;
>> };
>> #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, .left = NULL, \
>> - .count = 0, .min_vdisktime = 0, }
>> + .count = 0, .min_vdisktime = 0, .last_end_request = 0, \
>> + .ttime_total = 0, .ttime_samples = 0, .ttime_mean = 0 }
>>
>> /*
>> * Per process-grouping structure
>> @@ -962,8 +967,10 @@ cfq_find_alloc_cfqg(struct cfq_data *cfqd, struct cgroup *cgroup, int create)
>> goto done;
>>
>> cfqg->weight = blkcg->weight;
>> - for_each_cfqg_st(cfqg, i, j, st)
>> + for_each_cfqg_st(cfqg, i, j, st) {
>> *st = CFQ_RB_ROOT;
>> + st->last_end_request = jiffies;
>> + }
>> RB_CLEAR_NODE(&cfqg->rb_node);
>>
>> /*
>> @@ -1795,9 +1802,12 @@ static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq)
>>
>> /*
>> * Otherwise, we do only if they are the last ones
>> - * in their service tree.
>> + * in their service tree and the average think time is
>> + * less than the slice length.
>> */
>> - if (service_tree->count == 1 && cfq_cfqq_sync(cfqq))
>> + if (service_tree->count == 1 && cfq_cfqq_sync(cfqq) &&
>> + (!sample_valid(service_tree->ttime_samples ||
> Jeff,
>
> Are we closing sample_valid() bracket at right place here?
I think you know the answer to that. ;-) Thanks for catching this.
> I am wondering where it is helping you.
The idea behind the patch is to optimize the idling such that we don't
wait needlessly for more sync-noidle I/O when it likely isn't coming.
As mentioned in the patch description, it aims to bring the sync-noidle
workload, which is treated like a single cfqq, more in-line with the
handling of a single cfqq.
> If it is to bring in line with with sync tree (old implementation),
> then we should have also compared the think time with slice_idle?
I'm not sure I follow 100%. Are you saying we should disable idling for
the sync-noidle workload if the think time is too long? That sounds
reasonable.
> But comparing that here might not be the best thing as cfq_should_idle()
> is used in many contexts.
Again, looking for clarification on this point.
>> + cfqq->slice_end - jiffies < service_tree->ttime_mean)))
>> return 1;
>
> This comparision will also might break some logic in select_queue() where
> we wait for a queue/group to get busy even if queue's time slice has
> expired.
>
> ********************************************************************
> if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) {
> /*
> * If slice had not expired at the completion of last
> * request
> * we might not have turned on wait_busy flag. Don't
> * expire
> * the queue yet. Allow the group to get backlogged.
> *
> * The very fact that we have used the slice, that means
> * we
> * have been idling all along on this queue and it should
> * be
> * ok to wait for this request to complete.
> */
> if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list)
> && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) {
> cfqq = NULL;
> goto keep_queue;
> }
>
> *************************************************************************
>
> With this change, now above condition will never be true as
> cfq_should_idle() will always return false as slice has already expired.
> And that will affect group loosing its fair share.
>
> So I guess we can define new functions to check more conditions instead of
> putting it in cfq_should_idle()
Right, thanks for pointing this out. Do we have a test case that
exposes this issue? We really need to start a regression test suite for
CFQ. Also, I had promised to run some numbers for you with cgroups
enabled and I didn't. I'll get that data before the next posting.
Thanks for the review, Vivek!
-Jeff
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists