This post is more than 5 years old
22 Posts
0
1056
May 25th, 2011 14:00
Clariion CX4-240 Throughput Values
Hello-
We have a CX4-240 array that has recently had alot of ESX and non ESX hosts added to it via the fibre protocol. No iSCSI connections. I am starting to get alot of slowness complaints, but don't have a licensed version of Navishphere Analyzer to look at the data. I do have VKernel to monitor my vm performance and it is indicating I/O issues and high latency on my datastores. I have logged into some of my vm and non vm machines and have run Iometer to capture some baselines. Now, I am trying to find out from the user community what type of throughput I should expect. With Iometer on a system, using Access Specifications 32K, 0% Read, 0% random I am seeing 25 MB/s. This particular system has its own LUN and is using 4 GB fiber disks. When I run similar tests on systems with backend SATA disks, surpisingly my throughput is similar.
On another array with much less I/O using the same Iometer tool I am seeing 40 MB/s. Any thoughts on other ways I should be testing and/or what I should expect to see? I know if I call my rep they will tell my to buy FAST cache, but I don't have the $ right now....
As always, thank you for your thoughts.....
Erik
Storagesavvy
474 Posts
0
May 26th, 2011 11:00
Erik,
It’s very difficult to know what is happening within the array without looking at the performance data. Since you don’t have an Analyzer license you can’t look at the data yourself, but you can see some statistics… If you open the Properties windows for both SPA and SPB you can see the Dirty Pages and SP Utilization values. They will even update periodically. In order for this to work, Statistics Logging needs to be enabled in the Array Properties dialog.
Also, work with your EMC Sales team to pull a NAZ file (encrypted Analyzer archive) from your array and have them look at it.. Most of the TCs (Sales Engineers) have tools to view the NAZ data.
You mentioned you have quite a few hosts… There are several reasons you could see slow downs.
Cache Contention – ie: Forced Flushing will slow down the whole array even if it’s only caused by a subset of data.
Backend disks are too busy.. This will also cause forced flushing
Queue Full on the front end ports. If there are many hosts, and each host HBA is using a high value for Execution Throttle, you could exceed the queue depth on the array ports (max 1600 per port) which will cause a Queue Full condition. This pauses IO at the port briefly and hosts respond by setting their own queue depth to 1 for some period of time, gradually increasing it back to defaults.
All of these conditions can be seen with Analyzer data. I believe the Queue Full’s can be seen in ESX logs but I can’t remember the text of the message.
Richard J Anderson
Julien_LECORRE
92 Posts
0
May 25th, 2011 14:00
Hi Erik,
It is hard to tell you what you can expect in throughput because it depends a lot of your disk topology !
First you have 3 kind of parameters when you talk about performance:
Each of this three parameters can result in bad performance on your storage. Basically when you get more disks in your Pool/Raid Group you have better Throuhgput and Bandwith.you have also the RAID type use on your Pool/Raid that impact this parameters.
You should expect approximately the same throughput with sequential data for approximately the same count of drive in both SATA and FC environnement.
You could see without Analyzer licence th real time performance on your array.
We can help you a little bit more if you provide your test configuration [RAID Type, Number of disk, Number of VM on your DataStore etc....]
If you want more advice, EMC have Speed Gurus [and great tools
] who can help you in resolving your performance issues.
jps00
2 Intern
•
392 Posts
0
May 26th, 2011 04:00
Erik,
The likely reason you're seeing the same metrics for both SATA and FC drives, is that you're measuring the fully cached performance of your storage system.
If it were me and I wanted to attack this problem, I'd check the SAN first. I'd start by verifying that my network performance has not degraded as a result of the expansion. If anything, it removes the network from the variables being considered.
EMC CLARiiON Best Practices for Performance and Availablity, FLARE 30.0contains advice on how to get the most performance from your storage system. Best Practices is available on Powerlink.
elarso
22 Posts
0
May 26th, 2011 08:00
Thank you for the information. When you say I am measuring the fully cached performance do you mean I am seeing similar speeds for both types of drives because my write cache is the bottleneck because it is saturated which is indicated by the high dirty pages? If yes, and nothing pans out with my network troubleshooting do you think I should look to adding the FAST cache?