Skip to main content
Start a Conversation

Unsolved

This post is more than 5 years old

GijiGeorge

5 Posts

61753

September 24th, 2010 05:00

SATA RAID 10 vs SAS 15K RAID 5

Hello
I wonder if anyone can share some info on peformance of a iSCSI based storage (EQL PS6010) using SATA RAID 10 vs SAS 15K RAID 5?

Thank you in advance :)

DELL_LanceB

49 Posts

September 24th, 2010 15:00

GijiGeorge -

Here are some links to some documents that might help.

PS6000XV – Performance Oriented 15,000 RPM SAS For more information: http://www.equallogic.com/products/Default.aspx?id=7897

PS6000X – Balanced Performance and Capacity 10,000 RPM SAS For more information: http://www.equallogic.com/products/Default.aspx?id=7889

PS6000E – Cost Effective SATA For more information: http://www.equallogic.com/products/default.aspx?id=7873



GijiGeorge

5 Posts

September 29th, 2010 06:00

Hi Lance
Thank you for the info.

Indeed SAS 15K (RAID 5) would perform faster than SATA (RAID 5). But I believe RAID 10 would elevate the performance of SATA array. I need some figures to validate that.

In my scenario, 15x SAS 15K (RAID 10) would be too expensive and overkill.

Regards

CWCertus1

1 Message

October 1st, 2010 04:00

This isn't a bad indicator. Takes a bit of playing with but will give you an idea.
If you have any IOPS figures to hand, may be worth working out your scenario and testing this to see if it is worthwhile and to see if they come out accurately.

http://wmarow.com/strcalc/

JOHNADCO

847 Posts

October 1st, 2010 10:00

I don't know. So we tried the RAID 10 experiement as we were interested in the write improvement from raid 10.

14 slow SATA drives in a disk group. (so 14 drives to play with)
RAID 5 produced around 1000 IOPS average on 50% random writes - some testing periods showed as low as 800 IOPS but most testing periods showed above1000 and often thegiven testing period showed in the upper 1200's.

Raid 10 - produced 750 IOPS average and much more consistantly. On the high side we would hit in the upper 800's for some testing periods and I don't think it ever went below 700 on the low side from some periods of testing.

So what should we conclude from our test? I'm not entirely sure, but I will continue to build disk groups of raid 5 with as many spindles as I can throw at it.

I am also not sure how my little test applies to the original question here. But we just completed the testing yesterday into this morning and I thought I would throw it out there,

GijiGeorge

5 Posts

October 28th, 2010 02:00

Thank you all for the inputs.

Based on the above info, to my surprise it seems RAID 10 does not offer any increase in performance!

JOHNADCO

847 Posts

October 29th, 2010 08:00

"Thank you all for the inputs.

Based on the above info, to my surprise it seems RAID 10 does not offer any increase in performance!
"
It does, but you have to add at least couple of spindles. You can't cut the IOPS spindlecount in 1/2. is all which is what you do with raid 10, you only get 1/2 of the spindles in the disk group worth of IOPS. You get more consitant write speeds indeed. It is decieving becasuse most the time the data is written to controller cache too so on the back side it can wait for the write to actually occure while the server is on to other writes.

My finding are contrary to popular wisdom on this, but we have tested and retested this on our MD3000i's.

View More

View All

No Events found!

Top