[OpenIndiana-discuss] ISCSI SSD Performance

[Ian Collins]

Sašo Kiselkov wrote:
> On 02/18/2013 05:18 AM, Grant Albitz wrote:
>> I would like to discuss one more item:
>>
>> Based on the writes below vs the reads it seems like I am able to get more data out of a w/s as apposed to a read per second. I may just be misunderstanding the results but the disks themselves are rated for higher read performance then write. I am just wondering if there is a zfs tunable I can play with. The disks seem have much more queued up during writes then reads as well.
>>
>>
>> time dd if=/dev/zero of=/PSC.Net/dd.tst bs=2048000 count=131027
>> r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>>      0.0    0.0    0.0    0.0  0.0  0.0    0.0    0.0   0   0 c1t24d0
>>      0.0    0.0    0.0    0.0  0.0  0.0    0.0    0.0   0   0 c2t0d0
>>      0.0 1057.2    0.0 89869.6  0.0  9.6    0.0    9.1   1  67 c1t0d0
>>      0.0 1057.2    0.0 89852.5  0.0  9.6    0.0    9.1   1  67 c1t1d0
>>      0.0 1057.4    0.0 89946.6  0.0  9.6    0.0    9.1   1  66 c1t2d0
>>      0.0 1057.0    0.0 89912.4  0.0  9.6    0.0    9.1   1  67 c1t3d0
>>      0.0 1038.0    0.0 87068.1  0.0  9.5    0.0    9.1   1  66 c1t4d0
>>      0.0 1037.2    0.0 87015.8  0.0  9.5    0.0    9.1   1  66 c1t5d0
>>      0.0 1037.0    0.0 87076.6  0.0  9.5    0.0    9.2   1  66 c1t6d0
>>
>> time dd if=/PSC.Net/dd.tst of=/dev/null bs=2048000
>>    r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>>      0.0    0.0    0.0    0.0  0.0  0.0    0.0    0.0   0   0 c1t24d0
>>      0.0    0.0    0.0    0.0  0.0  0.0    0.0    0.0   0   0 c2t0d0
>>    884.1    0.0 38373.3    0.0  0.0  0.3    0.0    0.4   1  13 c1t0d0
>>    873.7    0.0 38425.6    0.0  0.0  0.3    0.0    0.4   1  14 c1t1d0
>>    883.5    0.0 38399.9    0.0  0.0  0.3    0.0    0.4   1  13 c1t2d0
>>    875.1    0.0 38408.2    0.0  0.0  0.3    0.0    0.4   1  13 c1t3d0
>>    876.5    0.0 38528.1    0.0  0.0  0.3    0.0    0.4   1  13 c1t4d0
>>    880.7    0.0 38520.2    0.0  0.0  0.3    0.0    0.4   1  14 c1t5d0
>>    874.9    0.0 38510.4    0.0  0.0  0.3    0.0    0.4   1  13 c1t6d0
> That PERC H710 is probably really messing with your results here. First
> of all, the transparent compression of SSDs seems to be helping you out
> in the first instance. You'll need to circumvent that sucker (e.g. by
> writing compressed files).
>
> Next, I suggest you get rid of that H710 if possible, stick the H310 in
> and configure it in JBOD mode - we can try to resolve the driver issues,
> there's been some work in Illumos on this lately, so you might be able
> to build some custom driver modules (or I can do it for you).

I agree with Sas's comments on the 710.

If you have driver problems with a 310, try Solaris 11 as an experiment 
to verify performance, I have a number of Dells with 310 in JBOD mode 
running Solaris and they work well.

-- 
Ian.