Unsolved
This post is more than 5 years old
207 Posts
0
1218
Symmetrix DMX4 Performance
We have a large DMX4 with many hosts connected (Windows, AIX, Linux, Unisys, IBM Mainframe, VMware etc). We are adding a new mission critical AIX server to the DMX4. It will use around 5TB of space. It runs a database calle Progress and has many interfaces and many large file systems.
We will be doing a stress test of the new AIX system. We will simulate hundreds of users and then look at AIX - memory, CPU, Ethernet Activity etc. On the Storagae side I want to monitor the symm devices that make up the most important file systems.
I think I have 2 options to monitor symm performance. One is to run symstat at intervals and check IOPS and Kbs throughput. The other is to check Performance Manager which is running on my ECC console. This will show IOPS, Kbs, %hit, number of write pending tracks, response time etc.
I can collect all this data. Problem is I don't really know how to measure the performance as it pertains to this new AIX server. I may see lots of IOPS, Kbs on the new devices but what does this mean. Lots of throughput may mean that the system is doing good - or it may be dragging down my DMX for the other servers and application etc.
Any thoughts on how to interpet this info would be very helpful.
Quincy561
1.3K Posts
1
June 10th, 2011 11:00
Response time is one of the best measurements of storage performance. Good response time means it is not overutilized.
I would not use symstat for this, too hard to process. You could use it for quick glimpses into the performance.
You can use the ECC perf manager output, or turn that off and turn on the STP daemon collection on some host that has a recent solutions enabler installed on it. You can open the TTP files with the performance viewer that should come with ECC. Or get EMC to start a STP collection on the SP. If you don't have 10s of thousands of devices in your Symmetrix, you can go for lower than 5 minute collections, 2 minutes or even 1 minute are possible with smaller numbers of devices.
brad12341
207 Posts
0
June 10th, 2011 16:00
Currently the array has about 8000 devices.
Today I started getting familiar with ECC and Performance Manager. I went to Policy Definitions (Storage Agent for Symmetrix) and turned on WLA Revolving_00 with a 2 minute interval. Hopefully I will have something to look at when I get in the office on Monday. I also set up the Host Agent for AIX to gather stats on the new AIX host.
Looking at the metrics for the Symm - I think I will select the following
System
KB read per second
KB Written per second
number write pending tracks
Kbytes transfered per second
total IOs per second
Port
IOPS
throughput in Kbytes per second
% utilized
Devices (I put all the new AIX devices in a device group so I can easily add them to Perf Manager)
IOPS
Kbytes read per second
Kbytes written per second
write pending count
I know I will have lots of good metrics. Next step will be to find out what they mean. Lots of throughput may be good for this AIX host but may be drag down other hosts connected to this array.
Quincy561
1.3K Posts
0
June 14th, 2011 09:00
Safely assume the following for small random I/O: *
–Each 15 K FC disk can do 150 IOPS
–Each 10 K FC disk can do 120 IOPS
–Each 7200 SATA disk can do about 70 IOPS
–Each EFD disk can do 2,000+ IOPS
* With reasonable response time
brad12341
207 Posts
0
June 14th, 2011 09:00
I have several hundred drives on the DMX4. Most are Seagate with a product Revision T300R10. I'm guessing these are 10k drives. In general how many IOPS and KBPs should I expect per drive. We have many different type of apps with different read/write profiles - so all I am looking for is general numbers.
brad12341
207 Posts
0
June 14th, 2011 10:00
What about general numbers for KBps for each drive( I know it depends). I am most concerned about the 10k and 15k FC drives.
dynamox
1 Rookie
1 Rookie
•
20.4K Posts
0
June 16th, 2011 05:00
brad12341
207 Posts
0
June 16th, 2011 09:00
OK, I just want to make sure I understand the table above. It looks like a 15k FC drive will do 8MBs when block sizes are 64K. With 512K blocks then it jumps up to 32MBs. Is this correct. I have mostly 15k FC drives.
Looking at most of my devices in Performance Manager the average block sizes are around 2k - 8k. So my KBs through put must be lower than 8MBs. Maybe 5 MBs for each disk drive?
The biggest exception is the small collection of VMware volumes - many of them average more than 400K for both reads and writes.
In the past I was assuming around 10MBs for each 15k FC drive. Based on my small block sizes it looks like I should assume about 5MBs throughput for each drive. Does this sound correct?
Quincy561
1.3K Posts
0
June 16th, 2011 10:00
For small IOs, please don't think about MB/sec but IOPs/sec. If you then want to convert to MB/sec simply multiply the IOPs times the IO size.
Safely assume the following for small random I/O: *
–Each 15 K FC disk can do 150 IOPS
–Each 10 K FC disk can do 120 IOPS
–Each 7200 SATA disk can do about 70 IOPS
–Each EFD disk can do 2,000+ IOPS
* With reasonable response time