Re: [patch 0/7] Add TRIM support for raid linear/0/1/10

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]


2012/3/15 Holger Kiehl <Holger.Kiehl@xxxxxx>:
> On Thu, 15 Mar 2012, Shaohua Li wrote:
>
>> 2012/3/15 Holger Kiehl <Holger.Kiehl@xxxxxx>:
>>>
>>> On Wed, 14 Mar 2012, Shaohua Li wrote:
>>>
>>>> Maybe the discard runs slow with small size request in the disk.
>>>> please drop patch "blk: add plug for blkdev_issue_discard" and try
>>>> again.
>>>> Since
>>>> we can't do merge, the plug just introduces latency.
>>>>
>>> Tried again without the patch applied, but there is only a very small
>>> performance increase (520->600 agains 4000 fps without discard).
>>>
>>> The benchmark creates lots of small files (2 KiB) and deletes them again.
>>>
>>>
>>>> if it doesn't help, please capture a blktrace when you do the benchmark
>>>> and
>>>> send it to me.
>>>>
>>> Ok, I will do this tomorrow. Need some sleep :-)
>>>
>>> Thanks for your work on supporting discard in MD!
>>
>> I tried your benchmark, create 2000k 2k files and delete them and
>> follows a sync.
>> the discard runs pretty fast for both raid 0/1. So can't reproduce the
>> issue. I'm using
>> a fusionio card though. I'm afraid nothing I can do till get you blktrace.
>>
> The blktrace is a bit large so I have uploaded it to:
>
>   ftp://ftp.dwd.de/pub/afd/test/trim/trace
>
> This is while the benchmark was running. Just a reminder, md2 is
> /home under which the benchmark was running. And md2 is a raid0 of
> sda3, sdb3 and sdc3. While md1 is / and also raid0 of ada2, sdb2 and
> sdc2.
>
> There is also another blktarce when all files are deleted and note
> this is only part of it (10 min), it takes about 30 minutes to delete
> all. You can find this here:
>
>   ftp://ftp.dwd.de/pub/afd/test/trim/trace2
>
> Please tell me if you need more information or what else I can do to
> help find the problem.
Looks at the blktrace:
8,0    1    47871   116.769583185   870  A   D 46042912 + 96 <- (8,3) 30869280
  8,3    1    47872   116.769583560   870  Q   D 46042912 + 96 [jbd2/md2-8]
  8,3    1    47873   116.769584613   870  G   D 46042912 + 96 [jbd2/md2-8]
  8,3    1    47874   116.769585255   870  I   D 46042912 + 96 [jbd2/md2-8]
  8,3    1    47875   116.769585693   870  D   D 46042912 + 96 [jbd2/md2-8]
  8,3    1    47876   116.771985862     0  C   D 46042912 + 1 [0]
  8,0    1    47877   116.799571098   870  A   D 46040696 + 32 <- (8,3) 30867064
  8,3    1    47878   116.799571462   870  Q   D 46040696 + 32 [jbd2/md2-8]
  8,3    1    47879   116.799572459   870  G   D 46040696 + 32 [jbd2/md2-8]
  8,3    1    47880   116.799573176   870  I   D 46040696 + 32 [jbd2/md2-8]
  8,3    1    47881   116.799573637   870  D   D 46040696 + 32 [jbd2/md2-8]
  8,3    1    47882   116.801970911     0  C   D 46040696 + 1 [0]
  8,0    1    47883   116.801980623   870  A   D 46046568 + 88 <- (8,3) 30872936
  8,3    1    47884   116.801980957   870  Q   D 46046568 + 88 [jbd2/md2-8]
  8,3    1    47885   116.801981894   870  G   D 46046568 + 88 [jbd2/md2-8]
  8,3    1    47886   116.801982539   870  I   D 46046568 + 88 [jbd2/md2-8]
  8,3    1    47887   116.801982974   870  D   D 46046568 + 88 [jbd2/md2-8]
  8,3    1    47888   116.811997203     0  C   D 46046568 + 1 [0]
  8,0    1    47889   116.829566908   870  A   D 46040032 + 32 <- (8,3) 30866400
  8,3    1    47890   116.829567261   870  Q   D 46040032 + 32 [jbd2/md2-8]
  8,3    1    47891   116.829569154   870  G   D 46040032 + 32 [jbd2/md2-8]
  8,3    1    47892   116.829569901   870  I   D 46040032 + 32 [jbd2/md2-8]
  8,3    1    47893   116.829570366   870  D   D 46040032 + 32 [jbd2/md2-8]
  8,3    1    47894   116.831972370     0  C   D 46040032 + 1 [0]
  8,0    1    47895   116.846461610   870  A   D 46039728 + 8 <- (8,3) 30866096
  8,3    1    47896   116.846462008   870  Q   D 46039728 + 8 [jbd2/md2-8]
  8,3    1    47897   116.846462911   870  G   D 46039728 + 8 [jbd2/md2-8]
  8,3    1    47898   116.846463530   870  I   D 46039728 + 8 [jbd2/md2-8]
  8,3    1    47899   116.846463984   870  D   D 46039728 + 8 [jbd2/md2-8]
  8,3    1    47900   116.851970109     0  C   D 46039728 + 1 [0]

there are 5 discard requests, the discard request uses 2ms, 2ms, 10ms, 2ms, 5ms
(from dispatch to finish). this isn't fast definitely. And since
discard runs in jbd, slow
discard will impact other file operations, for example, when journal is full.

So looks like this isn't the fault of my patch. curious is why discard
is fast without md
in your test. Maybe the reason is the files in a new formatted
filesystem haven't
fragmentation, so discard size is big and discard request number is
small, so total
discard time is small too. while your md filesystem might have fragmentation, so
discard size is small and request number is big.

Thanks,
Shaohua
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html


[ATA RAID]     [Linux SCSI Target Infrastructure]     [Managing RAID on Linux]     [Linux IDE]     [Linux SCSI]     [Linux Hams]     [Device-Mapper]     [Kernel]     [Linux Books]     [Linux Admin]     [Linux Net]     [GFS]     [RPM]     [git]     [Photos]     [Yosemite Photos]     [Yosemite News]     [AMD 64]     [Linux Networking]

Add to Google Powered by Linux