HP EVA 4000 Performance Monitoring - Questions

[SIMON PARROTT]

we are examining the â Microsoft Performance Monitorâ output from the EVA data monitoring service on the SMA server in order to try to understand intermittent I/O transaction timeouts which are causing problems. We have some questions about the output from the monitor and the configuration of the eva 4000

1. Where do the queues whose depths are being displayed for each path actually reside â are they internal to the controllers?
2. Are the numbers we are seeing in this spreadsheet reasonable or do they indicate stress?
3. Is there any intelligence in the queue management, or is it just FIFO? In other words, is there any intelligence acting on queue entries which might schedule them out of the order in which they were received?
4. Would it be possible to get some sort of statement regarding the maximum performance one could expect from the controllers for different block sizes?
5. Can we configure the devices so that certain operations go to a specific controller while others go to the opposite controller? If this I possible, is it recommended by hp?

[Ivan Ferreira]

>>> 1. Where do the queues whose depths are being displayed for each path actually reside â  are they internal to the controllers?

The queues displayed are the controller port queues.

>>> 2. Are the numbers we are seeing in this spreadsheet reasonable or do they indicate stress?

Last information I got is that EVA can handle a queue depth up to 2048 for each port.

>>> 3. Is there any intelligence in the queue management, or is it just FIFO? In other words, is there any intelligence acting on queue entries which might schedule them out of the order in which they were received?

I owe you that.

>>> 4. Would it be possible to get some sort of statement regarding the maximum performance one could expect from the controllers for different block sizes?

Hard to say and depends of the environment. Stress test is the best option.

>>> 5. Can we configure the devices so that certain operations go to a specific controller while others go to the opposite controller? If this I possible, is it recommended by hp?

Yes. You can set a preferred path on both, the storage controller serving the disk, and the HBA path in a multipath host.

From the csv output, it looks like your controllers cpu utilization is normal, data % and cpu% are close which is good.

Also, it looks like you are using a pure "round robin" algorithm for load balancing, this can lead in a lot of "proxy I/O" which can reduce the performance.

Also, I can see some "spikes" up to 200 MB/s. This is the bandwidht a 2Gbps SAN can provide.

[SIMON PARROTT]

Ivan.

Re q3) Is there any documentation on queue scheduling?

Re q4) Is there an HP stress test utility?

Re) Is there any an HP document on setting the preferred controller path and hba path?

thanks Simon

[Ivan Ferreira · Accepted solution]

>>> Re q3) Is there any documentation on queue scheduling?

Again I owe you that

>>> Re q4) Is there an HP stress test utility?

Please see:

http://h20000.www2.hp.com/bizsupport/TechSupport/Document.jsp?objectID=lpg50460&locale=en_US

>>> Re) Is there any an HP document on setting the preferred controller path and hba path?

This is done in Command View EVA, selec the vdisk, presentation, and there you can set the controller preference.

From the host side, depends of the multipath software used.

Have a nice day.

[Ivan Ferreira]

Please see also the attached document.

[Ted Buis]

In response to question #4, your performance is typically likely to be limited by the number of disk that are in your configuration. The EVA4000 has the same controllers as the EVA6000. The EVA6000 can typical scale up in performance as disk are added close to the maximum of 112. Since the EVA4000 can have a maximum of 56 drives, it is the disks, not the controller that typically limit performance in real world applications. Your performance is also determined by the number of disks in your disk group, the speed of those disks and the RAID levels that you set for each LUN. The EVA QuickSpecs for the EVAx100, have performance maximums, but these are maximums that are tied to the cache performance and don't reflect the backend performance of the physical disks and their configuration. The nice thing about an EVA is that you can add disks and improve performance. If you are at the the maximum 56 drives in the EVA4000, you can (for a short time) still upgrade it to an EVA6000 by adding the license, loop switches and extra disk enclosures. Adding disks to the disk groups that have the LUNs which need a boost, will help as the EVA will automatically load balance to take advantage of the new spindles.

[SIMON PARROTT]

Ivan, Ted thanks both for your answers. Regards Simon