VA7400 performance problem

[Doug Murdoch]

We have 2 VA7400's that are accessed by a combination of NT 4.0 sp6 and HP N4000 servers through brocade switches. We are using veritas volume manager 2.7 for NT. When NT users access a directory or a file for the first time there is a delay of up to 30 seconds before data is returned. Looking at the packets being transferred it appears that when the client request the data from the server, the server responds 5 times with no data,the client repeats the request and the server then responds with the data. This problem appeared shortly after we upgraded Veritas Netbackup from version 3.41 to version 4.5 although I don't know if that is significant or not. The problem occurs on all of our NT servers on both VA7400's

any suggestions?

Thanks

[Vincent Fleming]

This does not sound like a VA7400 performance problem. It sounds like a Windoze Performance Problem!

If the data was not being returned by the disk array, the server would not be responding to clients with empty packets... it would not respond at all (ie: it would hang waiting on the disk I/O.)

Good luck!

[Leif Halvarsson_2]

Hi

Does this problem only apply to file systems located on the VA7400 or is it related to veritas volume manager, or perhaps general to all NT fileservers. We have a similar performance problem on our NT servers (browsing/acessing can be very slow) but this seems to be a general NT problem for us.

[Roger_22]

The above suggestions are good. If you still suspect the VA:

1) Make sure the array is not in a degraded operation. Verify that there are no warnings or errors. Examine the event log (logprn), many entries during the time period of the poor performance (and you are not physically changing the configuration) could indicate an underlying hardware problem. Contact hp if you suspect an issue.

2) Next, update to HP17. HP17 firmware has a couple of enhancements for sequential and high utilization workloads. Multiple servers connected to an array can easily generate a high demand.

3) If, after HP17 is loaded and you still have a performance problem, you need to examine the performance logs (armperf). The manpages describe each statistic gathered by the array. You need to verify that the workload is evenly distributed to the two Redundancy Groups in each array ??? you may be using only half of each array. The easiest way to accomplish this is to import the statistics into Excel using the COMMA option in armperf. Using the LUN output, add all the IO operations together for the LUNs on each RG. This should be as balanced as possible.

Write if you need more help.

Roger

[Eugeny Brychkov]

In addition to all replies I would like to add that:
- VA battery firmware should be 4.3;
- if using HP Autopath please use "preferred path" policy and set-up preferred path to LUNs created in RG1 (redundancy group 1) through VA C1 (controller 1) and for LUNs created in RG2 through C2. These are prefomance paths.
Eugeny

[Eugeny Brychkov]

And one more thing.... If NT servers have HP D8602A/B FC HBA installed please update driver to the latest (go to "drivers" and search for "D8602"). Be cautious - just after upgrade check with "computer management" if you have old driver removed. If it's still there go to device manager ad delete FC HBA device and then do "scan for HW changes" will install new driver correctly.
Eugeny