All of lore.kernel.org
 help / color / mirror / Atom feed
From: Spelic <spelic@shiftmail.org>
To: Jens Axboe <JAxboe@fusionio.com>
Cc: "fio@vger.kernel.org" <fio@vger.kernel.org>
Subject: Re: Again on IOPS higher than expected in randwrite 4k
Date: Wed, 5 Jan 2011 04:45:48 -0700	[thread overview]
Message-ID: <4D2459EC.1030308@shiftmail.org> (raw)
In-Reply-To: <4D21D8BE.1000403@fusionio.com>

On 01/03/2011 03:10 PM, Jens Axboe wrote:
> On 2011-01-03 12:26, Spelic wrote:
>
>> On 01/03/2011 12:05 PM, Jens Axboe wrote:
>>
>>> On 2011-01-02 05:12, Spelic wrote:
>>>
>> Oh I see.
>> But if I add fsync=1 I still get 300 IOPS per disk, or even 500 on
>> very short seeks, so again I'd say these disks are cheating.  Do you
>> agree?
>>
> Did you verify that the fsync gets turned into a flush with eg blktrace?
> If it indeed is, then yes your number seems too high for that disk. With
> a SYNC_CACHE after each write, not even NCQ should be helping you (since
> each request will effectively be sync).
>

I'm not sure... I'm no expert of blktrace... can you have a look?
I am pasting part of the output below here.
(kernel 2.6.36.2)

I was issuing FIO to a lvm linear volume over a md raid10, with 20
threads doing 4K random writes in an 80MB file direct=1 and sync=1, I am
seeing about 450 IOPS per drive.
I captured one drive here.
There are many requests with "S", sync, but the S never reaches the "D"
driver. Does that mean there are no flushes?
I thought that DM linear and MD and raid10 in kernel 2.6.36.2 were
passing barriers to the layer below, so I thought that would work, or
anyway that the OS would work around that with some other technique so
to provide a reliable fsync (which is very important for data
consistency, isn't it).

Filesystem is ext4 mounted with defaults and it did not say nobarrier:
dmesg
[141306.496251] EXT4-fs (dm-2): mounted filesystem with ordered data
mode. Opts: (null)
cat /proc/mounts
/dev/mapper/datavg1-try2 /mnt/tmp2 ext4
rw,relatime,barrier=1,stripe=512,data=ordered 0 0

Another problem is: how come I see [swapper] operating... I had
specified direct=1 in fio!
[random-write]
rw=randwrite
rwmixcycle=1    #old settings...
rwmixread=50   #ignored in randwrite
numjobs=20
blocksize=4k
size=80m
directory=/mnt/tmp2/fio-data
direct=1
fsync=1
;iodepth=100   #this is commented out
end_fsync=1



blktrace --> blkparse:

  66,144  5     5009     5.256475644     0  D   W 20323192 + 8 [swapper]
  66,144  7     3180     5.256676923  6134  U   N [fio] 17
  66,144  5     5010     5.256681988   701  A  WS 33072392 + 8 <-
(66,145) 33070344
  66,144  5     5011     5.256682210   701  Q  WS 33072392 + 8 [md14_raid10]
  66,144  5     5012     5.256682992   701  G  WS 33072392 + 8 [md14_raid10]
  66,144  5     5013     5.256683642   701  I   W 33072392 + 8 [md14_raid10]
  66,144  5     5014     5.256685095   701  U   N [md14_raid10] 18
  66,144  5     5015     5.256927083  6129  U   N [fio] 18
  66,144  7     3181     5.256933947   701  U   N [md14_raid10] 18
  66,144  5     5016     5.257087648     0  C   W 19603416 + 8 [0]
  66,144  5     5017     5.257092870     0  D   W 21862904 + 8 [swapper]
  66,144  5     5018     5.257238735  6116  U   N [fio] 17
  66,144  7     3182     5.257243081   701  A  WS 19605560 + 8 <-
(66,145) 19603512
  66,144  7     3183     5.257243235   701  Q  WS 19605560 + 8 [md14_raid10]
  66,144  7     3184     5.257243779   701  G  WS 19605560 + 8 [md14_raid10]
  66,144  7     3185     5.257244203   701  I   W 19605560 + 8 [md14_raid10]
  66,144  7     3186     5.257245545   701  U   N [md14_raid10] 18
  66,144  5     5019     5.257546967  6118  U   N [fio] 18
  66,144  7     3187     5.257552944   701  U   N [md14_raid10] 18
  66,144  5     5020     5.257663684     0  C   W 20323192 + 8 [0]
  66,144  5     5021     5.257668596     0  D   W 22581240 + 8 [swapper]
  66,144  7     3188     5.257816612  6117  U   N [fio] 17
  66,144  5     5022     5.257823723   701  U   N [md14_raid10] 17
  66,144  5     5023     5.258130728  6129  U   N [fio] 17
  66,144  7     3189     5.258137688   701  U   N [md14_raid10] 17
  66,144  5     5024     5.258242581     0  C   W 21862904 + 8 [0]
  66,144  5     5025     5.258247884     0  D   W 23433784 + 8 [swapper]
  66,144  5     5026     5.258398322  6119  U   N [fio] 16
  66,144  7     3190     5.258402511   701  A  WS 21913520 + 8 <-
(66,145) 21911472
  66,144  7     3191     5.258402650   701  Q  WS 21913520 + 8 [md14_raid10]
  66,144  7     3192     5.258403128   701  G  WS 21913520 + 8 [md14_raid10]
  66,144  7     3193     5.258403567   701  I   W 21913520 + 8 [md14_raid10]
  66,144  7     3194     5.258404755   701  U   N [md14_raid10] 17
  66,144  5     5027     5.258653030  6118  U   N [fio] 17
  66,144  7     3195     5.258657124   701  A  WS 21150720 + 8 <-
(66,145) 21148672
  66,144  7     3196     5.258657262   701  Q  WS 21150720 + 8 [md14_raid10]
  66,144  7     3197     5.258657659   701  G  WS 21150720 + 8 [md14_raid10]
  66,144  7     3198     5.258658017   701  I   W 21150720 + 8 [md14_raid10]
  66,144  7     3199     5.258659238   701  U   N [md14_raid10] 18
  66,144  5     5028     5.258769609     0  C   W 22581240 + 8 [0]
  66,144  5     5029     5.258775005     0  D   W 24115736 + 8 [swapper]
  66,144  7     3200     5.258927319  6120  U   N [fio] 17
  66,144  5     5030     5.258932228   701  A  WS 22591840 + 8 <-
(66,145) 22589792
  66,144  5     5031     5.258932562   701  Q  WS 22591840 + 8 [md14_raid10]
  66,144  5     5032     5.258933374   701  G  WS 22591840 + 8 [md14_raid10]
  66,144  5     5033     5.258933985   701  I   W 22591840 + 8 [md14_raid10]
  66,144  5     5034     5.258935275   701  U   N [md14_raid10] 18
  66,144  5     5035     5.259234667  6117  U   N [fio] 18
  66,144  7     3201     5.259241744   701  U   N [md14_raid10] 18
  66,144  5     5036     5.259377131     0  C   W 23433784 + 8 [0]
  66,144  5     5037     5.259382301     0  D   W 24833912 + 8 [swapper]
  66,144  5     5038     5.259536908  6121  U   N [fio] 17
  66,144  7     3202     5.259541161   701  A  WS 23405256 + 8 <-
(66,145) 23403208
  66,144  7     3203     5.259541335   701  Q  WS 23405256 + 8 [md14_raid10]
  66,144  7     3204     5.259541945   701  G  WS 23405256 + 8 [md14_raid10]
  66,144  7     3205     5.259542385   701  I   W 23405256 + 8 [md14_raid10]
  66,144  7     3206     5.259543684   701  U   N [md14_raid10] 18
  66,144  5     5039     5.259788494  6129  U   N [fio] 18
  66,144  7     3207     5.259794434   701  U   N [md14_raid10] 18
  66,144  5     5040     5.259858786     0  C   W 24115736 + 8 [0]
  66,144  5     5041     5.259863634     0  D   W 25651296 + 8 [swapper]
  66,144  7     3208     5.260001478  6122  U   N [fio] 17
  66,144  5     5042     5.260008880   701  U   N [md14_raid10] 17
  66,144  5     5043     5.260339045  6117  U   N [fio] 17
  66,144  7     3209     5.260346008   701  U   N [md14_raid10] 17
  66,144  5     5044     5.260485907     0  C   W 24833912 + 8 [0]
  66,144  5     5045     5.260490978     0  D   W 26391696 + 8 [swapper]
  66,144  5     5046     5.260643566  6123  U   N [fio] 16
  66,144  7     3210     5.260647837   701  A  WS 24809912 + 8 <-
(66,145) 24807864
  66,144  7     3211     5.260648008   701  Q  WS 24809912 + 8 [md14_raid10]
  66,144  7     3212     5.260648565   701  G  WS 24809912 + 8 [md14_raid10]
  66,144  7     3213     5.260648914   701  I   W 24809912 + 8 [md14_raid10]
  66,144  7     3214     5.260650315   701  U   N [md14_raid10] 17
  66,144  5     5047     5.260905234  6129  U   N [fio] 17
  66,144  7     3215     5.260910790   701  U   N [md14_raid10] 17
  66,144  5     5048     5.260967679     0  C   W 25651296 + 8 [0]
  66,144  5     5049     5.260973084     0  D   W 27054336 + 8 [swapper]
  66,144  7     3216     5.261122944  6124  U   N [fio] 16
  66,144  5     5050     5.261127859   701  A  WS 25649792 + 8 <-
(66,145) 25647744
  66,144  5     5051     5.261128084   701  Q  WS 25649792 + 8 [md14_raid10]
  66,144  5     5052     5.261128845   701  G  WS 25649792 + 8 [md14_raid10]
  66,144  5     5053     5.261129450   701  I   W 25649792 + 8 [md14_raid10]
  66,144  5     5054     5.261131041   701  U   N [md14_raid10] 17
  66,144  5     5055     5.261409090  6122  U   N [fio] 17
  66,144  7     3217     5.261413722   701  A  WS 24147272 + 8 <-
(66,145) 24145224
  66,144  7     3218     5.261414176   701  Q  WS 24147272 + 8 [md14_raid10]
  66,144  7     3219     5.261415006   701  G  WS 24147272 + 8 [md14_raid10]
  66,144  7     3220     5.261415614   701  I   W 24147272 + 8 [md14_raid10]
  66,144  7     3221     5.261417027   701  U   N [md14_raid10] 18
  66,144  5     5056     5.261520670     0  C   W 26391696 + 8 [0]
  66,144  5     5057     5.261525858     0  D   W 27879824 + 8 [swapper]
  66,144  5     5058     5.261678537  6125  U   N [fio] 17
  66,144  7     3222     5.261682847   701  A  WS 26366992 + 8 <-
(66,145) 26364944
  66,144  7     3223     5.261683021   701  Q  WS 26366992 + 8 [md14_raid10]
  66,144  7     3224     5.261683451   701  G  WS 26366992 + 8 [md14_raid10]
  66,144  7     3225     5.261683818   701  I   W 26366992 + 8 [md14_raid10]
  66,144  7     3226     5.261685265   701  U   N [md14_raid10] 18
  66,144  5     5059     5.261922669  6117  U   N [fio] 18
  66,144  7     3227     5.261926032   701  A  WS 20358736 + 8 <-
(66,145) 20356688
  66,144  7     3228     5.261926140   701  Q  WS 20358736 + 8 [md14_raid10]
  66,144  7     3229     5.261926489   701  G  WS 20358736 + 8 [md14_raid10]
  66,144  7     3230     5.261926748   701  I   W 20358736 + 8 [md14_raid10]
  66,144  7     3231     5.261927696   701  U   N [md14_raid10] 19
  66,144  5     5060     5.262033970     0  C   W 27054336 + 8 [0]
  66,144  5     5061     5.262038854     0  D   W 28595752 + 8 [swapper]
  66,144  7     3232     5.262194327  6126  U   N [fio] 18
  66,144  5     5062     5.262202033   701  U   N [md14_raid10] 18
  66,144  5     5063     5.262461921  6129  U   N [fio] 18
  66,144  7     3233     5.262469612   701  U   N [md14_raid10] 18
  66,144  5     5064     5.262623725     0  C   W 27879824 + 8 [0]
  66,144  5     5065     5.262628847     0  D   W 30164048 + 8 [swapper]
  66,144  5     5066     5.262829611  6127  U   N [fio] 17
  66,144  7     3234     5.262834041   701  A  WS 27938424 + 8 <-
(66,145) 27936376
  66,144  7     3235     5.262834192   701  Q  WS 27938424 + 8 [md14_raid10]
  66,144  7     3236     5.262834883   701  G  WS 27938424 + 8 [md14_raid10]
  66,144  7     3237     5.262835340   701  I   W 27938424 + 8 [md14_raid10]
  66,144  7     3238     5.262836673   701  U   N [md14_raid10] 18
  66,144  5     5067     5.263133683  6126  U   N [fio] 18
  66,144  7     3239     5.263137858   701  A  WS 27061872 + 8 <-
(66,145) 27059824
  66,144  7     3240     5.263137999   701  Q  WS 27061872 + 8 [md14_raid10]
  66,144  7     3241     5.263138426   701  G  WS 27061872 + 8 [md14_raid10]
  66,144  7     3242     5.263138826   701  I   W 27061872 + 8 [md14_raid10]
  66,144  7     3243     5.263139975   701  U   N [md14_raid10] 19
  66,144  5     5068     5.263253802     0  C   W 28595752 + 8 [0]
  66,144  5     5069     5.263259327     0  D   W 30872560 + 8 [swapper]
  66,144  7     3244     5.263456187  6128  U   N [fio] 18
  66,144  5     5070     5.263461225   701  A  WS 28609536 + 8 <-
(66,145) 28607488
  66,144  5     5071     5.263461496   701  Q  WS 28609536 + 8 [md14_raid10]
  66,144  5     5072     5.263462308   701  G  WS 28609536 + 8 [md14_raid10]
  66,144  5     5073     5.263462876   701  I   W 28609536 + 8 [md14_raid10]
  66,144  5     5074     5.263464227   701  U   N [md14_raid10] 19
  66,144  5     5075     5.263610357  6129  U   N [fio] 19
  66,144  7     3245     5.263617058   701  U   N [md14_raid10] 19
  66,144  5     5076     5.263726740     0  C   W 30164048 + 8 [0]
  66,144  5     5077     5.263731763     0  D   W 31648664 + 8 [swapper]
  66,144  5     5078     5.263910048  6130  U   N [fio] 18
  66,144  7     3246     5.263916821   701  U   N [md14_raid10] 18
  66,144  5     5079     5.264179263  6129  U   N [fio] 18
  66,144  7     3247     5.264185501   701  U   N [md14_raid10] 18
  66,144  5     5080     5.264291934     0  C   W 30872560 + 8 [0]
  66,144  5     5081     5.264296909     0  D   W 32393312 + 8 [swapper]
  66,144  7     3248     5.264396840  6131  U   N [fio] 17
  66,144  5     5082     5.264404167   701  U   N [md14_raid10] 17
  66,144  5     5083     5.264712968  6130  U   N [fio] 17
  66,144  7     3249     5.264719768   701  U   N [md14_raid10] 17
  66,144  5     5084     5.264821254     0  C   W 31648664 + 8 [0]
  66,144  5     5085     5.264826506     0  D   W 33072392 + 8 [swapper]
  66,144  5     5086     5.265026755  6132  U   N [fio] 16
  66,144  7     3250     5.265032611   701  U   N [md14_raid10] 16
  66,144  5     5087     5.265294126  6129  U   N [fio] 16
  66,144  7     3251     5.265299691   701  U   N [md14_raid10] 16
  66,144  5     5088     5.265406558     0  C   W 32393312 + 8 [0]
  66,144  5     5089     5.265411617     0  D   W 18769952 + 8 [swapper]
  66,144  7     3252     5.265569929  6133  U   N [fio] 15
  66,144  5     5090     5.265577207   701  U   N [md14_raid10] 15
  66,144  5     5091     5.265923350  6131  U   N [fio] 15
  66,144  7     3253     5.265930496   701  U   N [md14_raid10] 15
  66,144  5     5092     5.266041020     0  C   W 33072392 + 8 [0]
  66,144  5     5093     5.266046233     0  D   W 19605560 + 8 [swapper]
  66,144  5     5094     5.266200776  6134  U   N [fio] 14
  66,144  7     3254     5.266206557   701  U   N [md14_raid10] 14
  66,144  5     5095     5.266454790  6130  U   N [fio] 14
  66,144  7     3255     5.266459012   701  A  WS 30133304 + 8 <-
(66,145) 30131256
  66,144  7     3256     5.266459253   701  Q  WS 30133304 + 8 [md14_raid10]
  66,144  7     3257     5.266459930   701  G  WS 30133304 + 8 [md14_raid10]
  66,144  7     3258     5.266460309   701  I   W 30133304 + 8 [md14_raid10]
  66,144  7     3259     5.266461473   701  U   N [md14_raid10] 15
  66,144  5     5096     5.266572863     0  C   W 18769952 + 8 [0]
  66,144  5     5097     5.266578124     0  D   W 20358736 + 8 [swapper]
  66,144  7     3260     5.266732219  6115  U   N [fio] 14
  66,144  5     5098     5.266737163   701  A  WS 18758136 + 8 <-
(66,145) 18756088
  66,144  5     5099     5.266737386   701  Q  WS 18758136 + 8 [md14_raid10]
  66,144  5     5100     5.266738168   701  G  WS 18758136 + 8 [md14_raid10]
  66,144  5     5101     5.266738776   701  I   W 18758136 + 8 [md14_raid10]
  66,144  5     5102     5.266740268   701  U   N [md14_raid10] 15
  66,144  5     5103     5.266997433  6132  U   N [fio] 15
  66,144  7     3261     5.267004469   701  U   N [md14_raid10] 15
  66,144  5     5104     5.267107097     0  C   W 19605560 + 8 [0]
  66,144  5     5105     5.267112123     0  D   W 21150720 + 8 [swapper]
  66,144  5     5106     5.267261824  6116  U   N [fio] 14
  66,144  7     3262     5.267267689   701  U   N [md14_raid10] 14
  66,144  5     5107     5.267565634  6133  U   N [fio] 14

Thank you



      reply	other threads:[~2011-01-05 11:45 UTC|newest]

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2011-01-02  4:12 Again on IOPS higher than expected in randwrite 4k Spelic
2011-01-03 11:05 ` Jens Axboe
2011-01-03 11:26   ` Spelic
2011-01-03 14:10     ` Jens Axboe
2011-01-05 11:45       ` Spelic [this message]

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=4D2459EC.1030308@shiftmail.org \
    --to=spelic@shiftmail.org \
    --cc=JAxboe@fusionio.com \
    --cc=fio@vger.kernel.org \
    /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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.