This post is more than 5 years old
61 Posts
0
2428
How many IOPS can i get out of RAID 5 metaLUN?
Hi.
I have a metalun configured on cx4-480 array. How many total throughput (IOPS) can I expect on the LUN?
There are 6 component LUNs on different RGs R5(8+1) combination. MetaLUN is striped NOT concatenated.
Read:60% Write: 40%
Please suggest. Also if you can show how did you calculate the total IOPS that will be just awesume.
Many thanks!!!
Regards,
Nishant Kohli
zhouzengchao
1.4K Posts
0
September 12th, 2013 19:00
If you are talking about IOPS, then I would suppose the i/o pattern is [random, small], maybe 4KB. But you didn't give out the disk type, I would use SAS 15krpm drives as an example.
let's do the calculation:
=================================
SAS 15krpm drive supports 180 IOPS
You have 6*8 = 48 data disks -> 48*180 = 8640 IOPS.
Another key point is that you need to consider RAID 5 write penalty (1 app write = 4 disk writes) when calculating the final disk level I/O to compare with the the value we got just now (8640).
Le't assume your application level IOPS is = X, then:
60%X + 40%X * 4 must <= 8640 -> x<= 3927, which means your application level I/O shouldn't be greater than 3927 with R/W cache disabled.
In summary, if we don't consider the cache i/o absorption, for i/o pattern = [random, small], 60%R | 40%W, with 15krpm SAS, you can sustain application i/o <=3927 IOPS.
Again, this value is just a result from a rule of thumb, EMC professional services can give you an more accurate value through some of the internal tools.
hope this helpful!
o17Uu33DCF12520
1.1K Posts
0
September 12th, 2013 19:00
In cx4-480, they are 15 rpm FC drives with 180 iops, the iops of RG(also of the component LUN) is 9*180=1620, the iops of metaLUN is 6*1620=9720.
btw, it's good to check out the metaLUN iops from the Analyzer statistics, and they should be accurate.
Nishant_Kohli
61 Posts
0
September 12th, 2013 21:00
Hey Steve, that really opened up some clogs in my mind. This is what I needed to know.
I forgot to mention the drive type, though you were right on 180 IOPS per drive. Mine is 15k FC.
Thank you so much for your help!
Regards
Nishant Kohli
Storage - Guitar Center, ITO | Desk: (818) 735-8800 ext 2835
Mobile – 818 584 5956
E: Nishant.Kohli@Guitarcenter.com
Nishant_Kohli
61 Posts
0
September 12th, 2013 21:00
Thanks Kevin! Appreciate your help.
Regards
Nishant Kohli
Storage - Guitar Center, ITO | Desk: (818) 735-8800 ext 2835
Mobile – 818 584 5956
E: Nishant.Kohli@Guitarcenter.com
o17Uu33DCF12520
1.1K Posts
0
September 12th, 2013 21:00
You are welcome.
Please mark my answer as "correct/helpful" answer if it helps. Thanks.
zhouzengchao
1.4K Posts
0
September 12th, 2013 22:00
glad to help!
christopher_ime
2K Posts
1
September 13th, 2013 21:00
Remember, when calculating the back-end IOPs, we should take in consideration *all* drives and should not be deducting the (equivalent of) one drive for the distributed parity of RAID 5. From a back-end perspective when considering IOPs (and also Bandwidth) we are calculating raw numbers regardless of what is written be it data or (distributed) parity. On the other hand, the write penalty of RAID 5 is factored in when calculating the front-end IOPs as you later noted.
Therefore, for a RAID 5 (8+1) RG, we should be calculating: 180 IOPs * 9 (not 8 just because those are the data disks).
zhouzengchao
1.4K Posts
0
September 14th, 2013 03:00
you are correct, thanks for the reminder.
RRR
5.7K Posts
0
September 18th, 2013 01:00
Also consider the fact that response times increase exponentially when drives are utilized over ±66%. This is called Little's law. Below ±66% response times are good, but almost every tool will show disk utilization over 66% in a red color. Yes you can push them to 100% utilization, but response times are severe. So if you need to keep some decent response times, stick with 66% of what you just calculated.
kelleg
4.5K Posts
0
September 20th, 2013 13:00
Lastly, the actual configuration of the metaLUN will depend on the other LUNs in the raid groups you use. If you configure correctly and stick to the best practices, you should get the optimal performance. See the attached for the guidelines for best configurations.
glen
1 Attachment
White Paper EMC CLARiiON MetaLUNs - A Detailed Review.pdf
zhouzengchao
1.4K Posts
0
September 22nd, 2013 20:00
right, RRR. We should multiply 70% (I always consider UT < 70% and its my first time get to 66% from you
) to make sure the UT is always under warning line. I think the rule of thumb of IOPS value say 180IOPS for SAS/FC 15krpm is concluded by running those disks goes up to 100% UT, isn't it?
RRR
5.7K Posts
0
September 23rd, 2013 03:00
The rule of thumb is 100% utilization, but the numbers are on the low / safe side. But still, consider 66% or 70% or something in that area as the tipping point / theshold for the noticable performance impact.
kelleg
4.5K Posts
1
September 24th, 2013 07:00
Was your question answered correctly? If so, please remember to mark your question Answered when you get the correct answer and award points to the person providing the answer. This helps others searching for a similar issue.
glen