raid3 is slow

Eric Anderson anderson at freebsd.org
Wed Mar 28 13:42:52 UTC 2007


On 03/28/07 08:21, R. B. Riddick wrote:
> --- Szabó Péter <matyee at mail.alba.hu> wrote:
>> raid3
>> Transfer rates:
>>         outside:       102400 kbytes in   1.179933 sec =    86785 kbytes/sec
>>         middle:        102400 kbytes in   1.148816 sec =    89135 kbytes/sec
>>         inside:        102400 kbytes in   1.171987 sec =    87373 kbytes/sec
>>
> Looks good...

It's about right, I agree.

>> #diskinfo -vt /dev/label/nmivol1
>>         outside:       102400 kbytes in   1.839020 sec =    55682 kbytes/sec
>>         middle:        102400 kbytes in   2.112483 sec =    48474 kbytes/sec
>>         inside:        102400 kbytes in   3.416896 sec =    29969 kbytes/sec
>>
> Looks slow...

You think so?  I think that's pretty acceptable for a UDMA100 device.

>> # atacontrol mode ad0
>> current mode = UDMA100
>> # atacontrol mode ad1
>> current mode = UDMA100
>> # atacontrol mode ad2
>> current mode = UDMA100
>> # atacontrol mode ad3
>> current mode = UDMA100
>> # atacontrol mode ad6
>> current mode = SATA150
>>
> Looks good...
> 
>> gstripe may help, and extends disk space, but less secure.
>>
> Yup...
> 
>> #dd if=/dev/zero of=/mnt/new/x.x bs=1m count=128
>> 128+0 records in
>> 128+0 records out
>> 134217728 bytes transferred in 11.410957 secs (11762180 bytes/sec)
>>
> Hmm


That looks pretty poor to me.  Can you send the output of a df -ik /mnt?


> And what if u try this:
>   dd if=/dev/raid3/... of=/dev/null bs=1m count=128
> ?
> 
>>> Disks   ad0   ad1   ad2   ad3   ad6                       intrn
>>> KB/t   8.42  8.42  8.42  8.42  5.67                113904 buf
>>> tps     287   287   287   287  1780                   133 dirtybuf
>>> MB/s   2.36  2.36  2.36  2.36  9.85                 69984 desiredvnodes
>>> % busy   25    26    24    26    99                 21586 numvnodes
> 8KB/transfer is a little low for my taste... Looks like ur copy program uses
> 32KB transfers, instead of 128KB transfers...
> 
>>> I use mc to copy.
> What if u try
>   dd if=/source-file/on/raid3-device of=/dev/null bs=1m count=128
> ?


It seems that the write speed is the real issue possibly.

A great test would be to write to the device itself, but that would 
blast the filesystem.  If that's ok, this would be a good test:

dd if=/dev/zero of=/dev/raid3/... bs=1m count=512


Eric




More information about the freebsd-geom mailing list