From: John Garry <john.garry@huawei.com>
To: Kashyap Desai <kashyap.desai@broadcom.com>, <axboe@kernel.dk>,
<jejb@linux.ibm.com>, <martin.petersen@oracle.com>,
<ming.lei@redhat.com>, <bvanassche@acm.org>, <hare@suse.de>,
<don.brace@microsemi.com>,
Sumit Saxena <sumit.saxena@broadcom.com>, <hch@infradead.org>,
Shivasharan Srikanteshwara
<shivasharan.srikanteshwara@broadcom.com>
Cc: <chenxiang66@hisilicon.com>, <linux-block@vger.kernel.org>,
<linux-scsi@vger.kernel.org>, <esc.storagedev@microsemi.com>,
Hannes Reinecke <hare@suse.com>
Subject: Re: [PATCH RFC v6 08/10] megaraid_sas: switch fusion adapters to MQ
Date: Wed, 8 Apr 2020 10:33:06 +0100 [thread overview]
Message-ID: <f839f040-8bf4-cf83-7670-dfc208b77326@huawei.com> (raw)
In-Reply-To: <a1f0399e2e85b2244a9ae40e4a2f1089@mail.gmail.com>
On 07/04/2020 12:14, Kashyap Desai wrote:
>> --- a/drivers/scsi/megaraid/megaraid_sas_fusion.c
>> +++ b/drivers/scsi/megaraid/megaraid_sas_fusion.c
>> @@ -373,24 +373,24 @@ megasas_get_msix_index(struct megasas_instance
>> *instance, {
>> int sdev_busy;
>>
>> - /* nr_hw_queue = 1 for MegaRAID */
>> - struct blk_mq_hw_ctx *hctx =
>> - scmd->device->request_queue->queue_hw_ctx[0];
>> + struct blk_mq_hw_ctx *hctx = scmd->request->mq_hctx;
>
Hi Kashyap,
>
> There is one outstanding patch which will eventually remove device_busy
> from sdev. To fix this interface, we may have to track per scsi device
> outstanding within a driver.
> For my testing I used below since we still have below interface available.
>
> sdev_busy = atomic_read(&scmd->device->device_busy);
So please confirm that this is your change in megasas_get_msix_index():
- sdev_busy = atomic_read(&hctx->nr_active);
+ sdev_busy = atomic_read(&scmd->device->device_busy);
>
> We have done some level of testing to know performance impact on SAS SSDs
> and HDD setup. Here is my finding -
> My testing used - Two socket Intel Skylake/Lewisburg/Purley
> Output of numactl --hardware
>
> available: 2 nodes (0-1)
> node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 36 37 38 39 40 41
> 42 43 44 45 46 47 48 49 50 51 52 53
> node 0 size: 31820 MB
> node 0 free: 21958 MB
> node 1 cpus: 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 54 55
> 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71
> node 1 size: 32247 MB
> node 1 free: 21068 MB
> node distances:
> node 0 1
> 0: 10 21
> 1: 21 10
>
>
> 64 HDD setup -
>
> With higher QD
what's OD?
> and io schedulder = mq-deadline, shared host tag is not
> scaling well. If I use ioscheduler = none, I can see consistent 2.0M IOPs.
> This issue is seen only with RFC. Without RFC mq-deadline scales up to
> 2.0M IOPS.
I didn't try any scheduler. I can have a look at that.
>
> Perf Top result of RFC - (IOPS = 1.4M IOPS)
>
> 78.20% [kernel] [k] native_queued_spin_lock_slowpath
> 1.46% [kernel] [k] sbitmap_any_bit_set
> 1.14% [kernel] [k] blk_mq_run_hw_queue
> 0.90% [kernel] [k] _mix_pool_bytes
> 0.63% [kernel] [k] _raw_spin_lock
> 0.57% [kernel] [k] blk_mq_run_hw_queues
> 0.56% [megaraid_sas] [k] complete_cmd_fusion
> 0.54% [megaraid_sas] [k] megasas_build_and_issue_cmd_fusion
> 0.50% [kernel] [k] dd_has_work
> 0.38% [kernel] [k] _raw_spin_lock_irqsave
> 0.36% [kernel] [k] gup_pgd_range
> 0.35% [megaraid_sas] [k] megasas_build_ldio_fusion
> 0.31% [kernel] [k] io_submit_one
> 0.29% [kernel] [k] hctx_lock
> 0.26% [kernel] [k] try_to_grab_pending
> 0.24% [kernel] [k] scsi_queue_rq
> 0.22% fio [.] __fio_gettime
> 0.22% [kernel] [k] insert_work
> 0.20% [kernel] [k] native_irq_return_iret
>
> Perf top without RFC driver - (IOPS = 2.0 M IOPS)
>
> 58.40% [kernel] [k] native_queued_spin_lock_slowpath
> 2.06% [kernel] [k] _mix_pool_bytes
> 1.38% [kernel] [k] _raw_spin_lock_irqsave
> 0.97% [kernel] [k] _raw_spin_lock
> 0.91% [kernel] [k] scsi_queue_rq
> 0.82% [kernel] [k] __sbq_wake_up
> 0.77% [kernel] [k] _raw_spin_unlock_irqrestore
> 0.74% [kernel] [k] scsi_mq_get_budget
> 0.61% [kernel] [k] gup_pgd_range
> 0.58% [kernel] [k] aio_complete_rw
> 0.52% [kernel] [k] elv_rb_add
> 0.50% [kernel] [k] llist_add_batch
> 0.50% [kernel] [k] native_irq_return_iret
> 0.48% [kernel] [k] blk_rq_map_sg
> 0.48% fio [.] __fio_gettime
> 0.47% [kernel] [k] blk_mq_get_tag
> 0.44% [kernel] [k] blk_mq_dispatch_rq_list
> 0.40% fio [.] io_u_queued_complete
> 0.39% fio [.] get_io_u
>
>
> If you want me to test any top up patch, please let me know. BTW, we also
> wants to provide module parameter for user to switch back to older
> nr_hw_queue = 1 mode. I will work on that part.
ok, but I would just like to reiterate the point that you will not see
the full benefit of blk-mq draining hw queues for cpu hotplug since you
hide hw queues from blk-mq.
>
> 24 SSD setup -
>
> I am able to see performance using RFC and without RFC is almost same.
> There is one specific drop, but that is generic kernel issue. Not related
> to RFC.
> We can discuss this issue separately. -
>
> 5.6 kernel is not able to scale very well if there is heavy outstanding
> from application.
> Example -
> 24 SSD setup and BS = 8K QD = 128 gives 1.73M IOPs which is h/w max, but
> at QD = 256 it gives 1.4M IOPs. It looks like there are some overhead of
> finding free tags at sdev or shost level which leads drops in IOPs.
>
Thanks for testing,
John
>
next prev parent reply other threads:[~2020-04-08 9:33 UTC|newest]
Thread overview: 35+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-03-05 11:54 [PATCH RFC v6 00/10] blk-mq/scsi: Provide hostwide shared tags for SCSI HBAs John Garry
2020-03-05 11:54 ` [PATCH RFC v6 01/10] blk-mq: rename BLK_MQ_F_TAG_SHARED as BLK_MQ_F_TAG_QUEUE_SHARED John Garry
2020-03-05 11:54 ` [PATCH RFC v6 02/10] blk-mq: rename blk_mq_update_tag_set_depth() John Garry
2020-03-05 11:54 ` [PATCH RFC v6 03/10] blk-mq: Use pointers for blk_mq_tags bitmap tags John Garry
2020-03-05 12:42 ` Hannes Reinecke
2020-03-05 11:54 ` [PATCH RFC v6 04/10] blk-mq: Facilitate a shared sbitmap per tagset John Garry
2020-03-05 12:49 ` Hannes Reinecke
2020-03-05 13:52 ` John Garry
2020-03-05 11:54 ` [PATCH RFC v6 05/10] blk-mq: Add support in hctx_tags_bitmap_show() for a shared sbitmap John Garry
2020-03-05 12:52 ` Hannes Reinecke
2020-03-05 11:54 ` [PATCH RFC v6 06/10] scsi: Add template flag 'host_tagset' John Garry
2020-03-06 11:12 ` John Garry
2020-03-05 11:54 ` [PATCH RFC v6 07/10] scsi: hisi_sas: Switch v3 hw to MQ John Garry
2020-03-05 12:52 ` Hannes Reinecke
2020-03-05 11:54 ` [PATCH RFC v6 08/10] megaraid_sas: switch fusion adapters " John Garry
2020-04-07 11:14 ` Kashyap Desai
2020-04-08 9:33 ` John Garry [this message]
2020-04-08 9:59 ` Kashyap Desai
2020-04-17 16:46 ` John Garry
2020-04-20 17:47 ` Kashyap Desai
2020-04-21 12:35 ` John Garry
2020-04-22 18:59 ` Kashyap Desai
2020-04-22 21:28 ` John Garry
2020-04-23 16:31 ` John Garry
2020-04-24 16:31 ` Kashyap Desai
2020-04-27 17:06 ` John Garry
2020-04-27 18:58 ` Kashyap Desai
2020-04-28 15:55 ` John Garry
2020-04-29 11:29 ` John Garry
2020-04-29 15:50 ` Kashyap Desai
2020-04-29 17:55 ` John Garry
2020-04-30 17:40 ` John Garry
2020-04-30 19:18 ` Kashyap Desai
2020-03-05 11:54 ` [PATCH RFC v6 09/10] smartpqi: enable host tagset John Garry
2020-03-05 11:54 ` [PATCH RFC v6 10/10] hpsa: enable host_tagset and switch to MQ John Garry
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=f839f040-8bf4-cf83-7670-dfc208b77326@huawei.com \
--to=john.garry@huawei.com \
--cc=axboe@kernel.dk \
--cc=bvanassche@acm.org \
--cc=chenxiang66@hisilicon.com \
--cc=don.brace@microsemi.com \
--cc=esc.storagedev@microsemi.com \
--cc=hare@suse.com \
--cc=hare@suse.de \
--cc=hch@infradead.org \
--cc=jejb@linux.ibm.com \
--cc=kashyap.desai@broadcom.com \
--cc=linux-block@vger.kernel.org \
--cc=linux-scsi@vger.kernel.org \
--cc=martin.petersen@oracle.com \
--cc=ming.lei@redhat.com \
--cc=shivasharan.srikanteshwara@broadcom.com \
--cc=sumit.saxena@broadcom.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).