VA-7100 slow performance

[Kay Liesenfeld]

Thank you for the info.

Unfortunalety the volume groups and logical volumes are created with Veritas Volume Manager and not visible for vginfo, vgreduce and vgextend.

I'll check after the weekend if the same things can be made with Veritas.

But, what I do not understand: I can't create LUNs on the controller I wish to. I can create a LUN and that's it.

I attached my ioscan for detailed info.

What you see is: I have created six LUNs without having the possibility to put one LUN on one controller. It's all made by the system.

As I said before, VA7100 only have one Redundancy Group, if you mean that.

But I'm still learning, so tell me please! ;-)

[David Bell_1]

Kay,

In Command View SDM, you cannot specify the controller in which a LUN is created on. The LUN will be created on the primary controller. That is why the suggestion of the vgreduce/vgextend. You are correct in that the VA7100 only has one RAID group. I'm not sure how that would be accomplished with Volume Manager, however, I'm sure that it has a way to control which path is primary to the Volume Group. I'll see if I can find anything on making those changes.

Dave

[Vincent Fleming]

With the VA7100, controller 1 is primary for all LUNs.

Good luck!

[Kay Liesenfeld]

I'd make a call at HP today and maybe they can help. There will be someone who will flash the firmware tomorrow, however.

Meanwhile, I did some tests with a 600MB-file:

dd if=file.dmp of=/dev/null

The VA needs 350 seconds for 600 MB = 1,6 MB per second. I read something about 90 MB in the factsheet. Sure, dd is no performance measurement tool, but the difference to theory is too large.

Using glance I mentioned that there is no i/o activity on the VA-devices, but very high cpu workload(about 90% for dd process). The file was for sure NOT in the buffer cache.

I have the same feeling with using Oracle and another i/o-intensive application on that machine: waiting for i/o and very high CPU workload.

I checked the inside performance data of the VA and the highest peak was at 20 Units / second. (MB???)

We connected a very old RAID5-Array (5 years, very slow!) and this array is faster (!!!) than the VA! Only difference: the old array was connected over LVD, not FC like the VA.

Maybe there's a bug in FC-handling of HPUX / driver?

Any idea by thinking that way? Any help is truly welcome!

[Kay Liesenfeld]

Hi all,

got the problem!

Example: 4 GB File
dd if=file of=/dev/null bs=512
350 seconds = 1,6 MB/s

dd if=file of=/dev/null bs=100M
72 seconds = 55 MB/s

If the application is getting larger blocks at one time and doesn't has to start / continue reading often, it's much better than with smaller block sizes.

OK. But how have have I to tune my applications to profit of that behaviour?

We are using vxfs, do you have any suggestions about cache size, extent size or something else?

And are there any ideas how to tune Oracle that way?

Thanks for everything!