Unsolved
This post is more than 5 years old
211 Posts
1
24685
IOPS calculation
How do I calculate the average IOPS in a (2+2) RAID 10 environment over 4 15K disks.
According specs: max. 380 (avg. 190) I/Os per second ST3450857FC 15K.7 450 GB
Is it 190x4=760 IOPS? What is my writepenalty in this configuration?
Am I right when I say 1+(parity/#hdd) = 1+(2/4) = 1,5 write penalty ???
DaveZ1
75 Posts
0
January 25th, 2011 09:00
yes you are correct,
Is the formula in a RAID5(4+1) environment:
r*h +(w*h*4) = diskIOPS
or
r*h +w*(h*4) = diskIOPS
or
r*h +(w*h)*4 = diskIOPS
due to operator ordering and Commutative Property of Multiplication the parentheses are irrelevant.
teovmy
211 Posts
0
January 25th, 2011 23:00
Thanks guy's for all your help.
RRR
2 Intern
2 Intern
•
5.7K Posts
0
January 26th, 2011 07:00
Example:
1000 Host IOps, Read/write ratio of 75%/25%
750 Read IO's and250 Write IO's
In RAID5 the penalty is 4 for random IO
Total number of IO's: 750 + 4 x 250 = 750 + 1000 = 1750 for your back end, so the disks should be capable of doing 1750 IOps.
If you have 15k disks in RAID5, you'd need 1750/180 = 9,7 = 10 disks, so this could for example be a META of 2 components in RAID5 (4+1)
In RAID1 this will ad up to:
750 + 2 x 250 = 1250 backend IOps.
1250 / 180 = 6 ,9 = 8 (in RAID10 you always have an even amount of disks), so this could be a RAID10 in 44 config or a META over 2 RG's of 22 each.
Seen from a capacity point of view the RAID5 solution offers the space of 2 x 4 disks if you choose the META solution of 2 x (41) = 8 disks or 9 disks if you created 1 RG in 91.
RAID10 offers 4 disks of space.
If the "RAID10 space" is sufficient, this is the cheaper solution for this IO profile.
DaveZ1
75 Posts
0
January 26th, 2011 07:00
good illustration, RRR, of the "25% rule" - I drag that out every time I talk of RAID selection. Of course these days power & cooling have to be factored in, which will give some economic advantage back to RAID 5.
RRR
2 Intern
2 Intern
•
5.7K Posts
0
January 26th, 2011 07:00
True: less disks are better and in that case even cheaper than just the cost of only the disks.
You will also have more disk slots available for future expansions, so you don't need to buy a new VNX any time soon....
SKT2
2 Intern
2 Intern
•
1.3K Posts
0
February 14th, 2011 03:00
This R5 advantage will be valid only when the R10 space is not sufficient . In the current illustration advantage goes with R10 holding the lesser number of disks.
Druehl1
223 Posts
0
December 12th, 2011 08:00
Hello,
one other question regarding IOPS.
Most of the cusotmer don´t know their IOPS or read write ratio.
So many of them want to know about the max. IOPS of a specialy disk configuration and say they have about 60/40 read/write.
Example:
1 x SAS Raid5 (6+1) + 2 x NL-SAS R6 (4+2): What are maximum the IOs of this configuration?
Is it:
1. Raid5: 7 x 180 = 1260 IOs
Raid6: 6 x 90 = 540 IOs
--> System has about 2340 IOPS
2. Raid5 r/w = 60/40: 0,6 x (7 x 180 IOs) + (0,4 x (7 x 180 IOs)) / 4 = 756 + 126= 882 IOs
Rad6 r/w = 60/40: 0,6 x (6 x 90 IOs) + (0,4 x (6 x 90 IOs)) / 6 = 324 + 36 = 360
--> System has about 1562 IOPS
I know EMC don´t give much informaton about the maximum IOPS and I have the Perf. and Avail. Guide in which you can calculate the systems when you have the needed IOPS, but as we all know the other way is the way we would need more often in the field.
DaveZ1
75 Posts
0
December 12th, 2011 10:00
#1 describes performance *potential* of the drives in question, but it's NOT max IOPS - it's Rule Of Thumb IOPS. It's a decent number to use for figuring small random IO. Too many variables needed to get "max" IOPS.
#2 is not a use approach. For #2 you need to solve for the host load that approaches the disk maxima. there's a 2 equations/2 unknowns approach (which I have solved in the past but cant find just yet) but you can determine if your answer is correct by putting it back through the equation in post 16. Take the RAID 5 equation:
882 * 0.6 + 4 * (882* 0.4) = 1940 IOPS. That is not your potential of 1260 IOPS for the RAID5 SAS set.
So your equation is incorrect.
kelleg
4.5K Posts
0
December 13th, 2011 13:00
Just as an example, I have seen one configuration where a set of 15K RPM FC disks in a 4+1 R5 were pushed to 450 IOPS for each disk. Response times for this particular LUN were around 15ms. This was a very special case and I've only seen this level of disk IO a couple of times over the last 8 years.
When trying to figure maximum's for anything, you can always find examples of someone that could push disks beyond what you would normally think possible. That's why it's important to understand that the numbers provided in the Best Practice guides are general guidelines, but they are not the maximum, just a good numbers to start from.
glen
Druehl1
223 Posts
0
December 13th, 2011 23:00
Hi,
thanks for your answers.
So lets assume I need average IOPS from a disk configuration I would onyl multiply the number of disks in the raid group with the average IOPS given in the guidelines and ignore the Raid-level at this time?
DaveZ1
75 Posts
0
December 14th, 2011 07:00
David,
If determining best fit WRT RAID type etc. follow the steps in the Best practice guide:
1) Determine performance potential of the drives; you can use a ROT IOPS rate if applying a small random workload, they will be close enough
2) Determine disk load from the host load by applying the read/write split & RAID factors to the writes
3) disk load / performance potential = disk saturation. Aim for 70%
If you have more questions please read the BPG carefully. thanks.
teovmy
211 Posts
0
January 24th, 2012 06:00
This has become a very interesting topic.
RRR
2 Intern
2 Intern
•
5.7K Posts
0
January 26th, 2012 03:00
Oh RLY ? Did you delete it from your mind again ?
kelleg
4.5K Posts
0
January 27th, 2012 14:00
ohhhh
danielsalud
16 Posts
0
August 27th, 2013 08:00
where do you get the document for IOPS calculation - EMC's version?