This post is more than 5 years old
7 Posts
0
3555
CX4 SSD Drives very high response times
Trying to run DB2 on SSD drives in a CX4-240. The drives are configured using 2 - RAID groups of RAID5 4 + 1 400GB SSD drives, which are allocated with 2 disk in one DAE and 3 in another and they are in BUS 0 Enclosure 0 and BUS 1 Enclosure 0. I have created 8 LUNS as META'S with 50GB in one RAID group and 50 in the other. The problem appears to be that DB2 is striping the IO across the LUNS and I also have these LUNS defined as striped META'S. The read and write ms response times are as high as 300 to 500 at times. I have tried all variations with cache both read and write enabled, just write, and none but this does not seem to make a difference. Would anyone have any suggestion on how to possibly improve this? Thanks
kelleg
4.5K Posts
0
April 27th, 2017 08:00
The below KB will provide more information about the recommended cache settings for the CX4. The process to change the Read and Write cache settings is pretty easy - at the end of this KB there is a description of the steps needed. The important point is that when you disable Write cache the overall performance on the array will decrease and this will impact latency on all applications using the array. This is why it's recommended to do this during a slow period. The whole process only takes about 60 seconds. You can also check in the Help section for more information about setting the System Cache. Just open the Properties window and click on the Help button - that will bring up the specific Help for the screen your on.
https://support.emc.com/kb/330735
glen
kelleg
4.5K Posts
0
March 29th, 2017 12:00
If you have Aanlyer installed on the array, try monitoring in Read-Time mode the front-end ports for the host that owns the metaLUN. The backend on the CX4 is 4Gb, so if your seeing higher Bandwidth on the front-end ports, you may be overloadiing the backend - 4Gb is 400MB/s and realistically more like 320MB/s before the bus gets overloaded.
For metaLUNs are you using one LUN from each Raid Group to create one metaLUN? You mentioned you created 8 LUN - did you create 8 LUNs in each Raid Group, then create each metaLUN using one LUN from each RG in the same order as in each RG?
White Paper EMC CLARiiON MetaLUNs - A Detailed Review
Glen
mgdtma
7 Posts
0
March 30th, 2017 09:00
Hello,
We do have analyzer installed and I will see if I can figure out how to monitor the front end ports.
On your Metal LUN question the answer is no. We did not have additional RAID groups with SSD in it however I do now.
I have 7- 400GB SSD drives in Bus 1 Enclosure 0 and 8 in Bus 0 Enclosure 0 and these make up 3 – RAID5 4 + 1 Raid Groups. The RAID groups consist of` 2 drives in one Bus and 3 drives in the other Bus. Like you mentioned I read somewhere you wanted to have the drives in both buses to increase from 320MB/S to 640MB/S.
The HBA’S are 8GB on the host but I have had EMC look at the Powerpath configuration and that all checked out as far as drivers and settings go.
What I see the most is High Queue Length for SPA. I have the SSD drives balanced between the two and some meta heads are on SPA and some on SPB.
So your suggestion is create 4 – 50GB luns will say and combine them all into one Meta Lun and have each one in a different Raid Group?
kelleg
4.5K Posts
0
April 4th, 2017 15:00
Yes - check out the Best Practice guide I attached for Stripped MetaLUNs - best configuration. You want to have one LUN from each raid group - four LUNs in total - that you you spread the workload over all the disks in each RG.
Next step would be to open a performance case with EMC to check the back end bandwidth - you'll need to have Data Logging running at 120 Seconds. See KB 473729 for a list of KB's about performance - how to run the data logging and how to examine the archives (called NAR files) that are collected.
glen
mgdtma
7 Posts
0
April 5th, 2017 03:00
Thanks, Glen
The CX4 is no longer supported by EMC and the third party support we have only includes hardware issues. Thanks for the responses
kelleg
4.5K Posts
0
April 7th, 2017 09:00
Another thought. The best practice for Databases recommended that you have two different types of Raid Groups (or Pools) for the Database and the Logs/Temp. DB files recommended using R5 with FAST Cache enabled (if using spinning disk). For LOGs/TMP, recommended R10 - two spinning disks 1+1 (no SSD) without FAST Cache.
glen
mgdtma
7 Posts
0
April 7th, 2017 10:00
The database and logs are on different volumes. The databases volumes are SSD on the CX4 and the log volume is on 15K local disk. I created 8 new LUNS which all have 25GB on each Raid Group. There are two Meta Heads on each Raid Group and 2 LUNS in each meta have SPA as the default and 2 have SPB. The results were still not good. We do not have FAST Cache and the actual cache in the CX4 is configured with 746mb read and 515mb write. The SSD LUNS have write cache enabled but not read.
LUN Name
Capacity
Throughput (IOPS)
Bandwidth (MB/s)
Utilization (%)
Service Time (ms)
Response Time (ms)
Queue Length
LUN 112
100.00
634.29
14.77
61.89
0.98
36.16
22.94
LUN 136
100.00
641.74
14.89
61.87
0.96
59.85
38.41
LUN 101
100.00
639.84
14.88
61.80
0.97
43.52
27.85
LUN 125
100.00
644.16
14.91
61.61
0.96
53.35
34.37
LUN 100
100.00
661.71
15.40
59.76
0.90
24.27
16.06
LUN 137
100.00
658.91
15.29
59.73
0.91
35.86
23.63
LUN 124
100.00
657.99
15.37
59.50
0.90
35.24
23.19
LUN 113
100.00
648.34
15.17
59.32
0.91
25.40
16.47
kelleg
4.5K Posts
0
April 10th, 2017 13:00
See if assigning more memory to Write cache helps - use 1000MB for Write and the rest to Read. There are very few configurations that need more Read cache than Write cache and the more Write cache you have generally results in overall better performance.
mgdtma
7 Posts
0
April 12th, 2017 05:00
Can this be changed on the fly?
Is there any doc on how this is done?
kelleg
4.5K Posts
0
April 12th, 2017 13:00
See KB 427088 https://support.emc.com/kb/427088 Setting VNX array Cache Page and HWM/LWM values to best practices values
glen
kelleg
4.5K Posts
0
April 26th, 2017 14:00
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
mgdtma
7 Posts
0
April 27th, 2017 03:00
Glen,
The recommendation from our business partner was to not change the cache settings on the CX4 without assistance. Having never done this I have decided to just leave it unless you have done it and there isn’t much to it? Thanks
mgdtma
7 Posts
0
April 27th, 2017 09:00
I apologize Glen first time I have used this site for a question. I selected Correct answer after your last reply and I am not sure where I give points. I saw 300 at the top so I click that a few times.