Unsolved
This post is more than 5 years old
2 Intern
•
222 Posts
0
1473
Lun performance issue
I have 2 DAE's filled with the CX-SA07-020 drives (30 drives total)
Ive got that setup as 2 disk groups, Raid-5, 14+1.
I creates a bunch of luns on both disk groups and presented to host.
On the host, I striped the luns using a stripe-width of 2, meaning that im only hitting a pair of luns (one from each disk group) at a time.
The performance is really poor.
Device: rrqm/s wrqm/s r/s w/s rMB/s wMB/s avgrq-sz avgqu-sz await svctm %util
dm-2 4264.80 0.00 634.80 0.00 307.36 0.00 991.62 133.45 207.95 1.58 100.00
dm-4 4240.40 0.00 659.00 0.00 308.89 0.00 959.94 114.14 173.42 1.52 100.00
dm-12 0.00 0.00 9780.60 0.00 610.15 0.00 127.76 1913.81 194.36 0.10 100.00
dm-2 and dm-4 are the individual luns, one from each DAE (and controller).
dm-12 is the stripe itself and is being used as a filesystem.
As the numbers show, the wait times are extreem, the utilization is 100% and the throughput (both read and write) are peaking around 600MB/sec.
I also tried a smaller number of disks in a disk-group, and that was worse. The 2 large disk groups yeilds the best performance, but it's well below what I was hoping for with 30 drives backing the store.
Would 4x DAE's with the 2Tb drives get me twice the throughput?
Any suggestions would be helpfull.
usao
2 Intern
2 Intern
•
222 Posts
0
July 22nd, 2016 08:00
Thanks, this is on a CX4-240, I have 2 DAE's with the 2TB SATA drives, one on each bus.
Given the math above, I should be happy im peaking around 600MB/s, as I should expect only half that.
The array itself has 4x 8GBit FC ports (2 on each SP), so I was thinking I could get about 3.2GByte/sec out from the array, but I wasn't expecting that much because I know the SATA is slower and only have 2 DAE's with SATA.
kelleg
4.5K Posts
0
July 22nd, 2016 08:00
Remember that the disks control the ultimate performance on the array. Also, with a 14+1 Raid 5 that if two disks fail you could lose all data (double faulted raid group). With SATA disks, EMC recommends using Raid 6 to protect against a two disk failure and the more disks you have in a raid group the higher the risk of more than one disk failing.
With Raid 6 there is also higher raid penalty compared to Raid 5. The White paper talks about this.
glen
kelleg
4.5K Posts
1
July 22nd, 2016 08:00
You didn't mention what model array you're using - it's helpful to know the array type as performance is different for each array and the different . The type of disk you're using is the 2TB SATA II disks. These are 7200 RPM disks and each disk can handle about 80 IOPS or about 8MB/s.
One simple way you can determine the performance capacity of the raid groups would be to multiple the number of disks in the raid group by the value you want to determine. If you want IOPS, then you would multiple 80 times the number of disks.
For example if you have a raid 5 with 5 disks (4+1) you would multiple 80 IOPS times 5 = 400 IOPS (this is approximate and is based on an IO size of 8KB, Random, 808/20 Read/Write).
If you use bandwidth, then use 8MB/s times 5 = 40MB/s.
For 30 disks, a rough estimate would be 30 * 80 = 2400 IOPS or 30 * 8 = 240MB/s
See the following PDF for a more in-depth explanation of configuring the array for best performance - see Chapter 5
EMC CLARiiON Performance and Availability Release 30 Firmware Update Applied Best Practices.pdf
glen
kelleg
4.5K Posts
0
July 22nd, 2016 09:00
It likely that the IO type is helping. I've seen the SATA II perform much better than expected due to the IO type the applications use - more threads, larger IO Size, more sequential, etc.
glen
usao
2 Intern
2 Intern
•
222 Posts
0
July 22nd, 2016 09:00
Thanks, I may change to a Raid6 (12+2) with 1 hot spare per DAE, and add 2 more DAE's of the 2Tb drives (assuming I can find them).
It's been stable for the past few months, only 1 disk failure, but the throughput was less than I expected. After looking over the PDF from you, I now see that im doing pretty well considering the limited number of disks.