Ask the Expert: Performance Calculations on Clariion/VNX

Thanks every one for participating and sharing ideas on this post and helping the community! I don't want to break current conversation about the logs but I would like to see some discussion on the perofrmance monitoring tools and use cases with most "important parameters" that needs to be considered for doing analysis in any environment. I understand that performance analysis is a ocean and varies from every scenario.

I know few of them:

Symmetrix - SPA

Clarrion - Navisphere analyzer/Unispere analyzer

VNX - Unispere analyzer

Celerra(NAS) - ?

VBLOCK (converged infrastructure) -?

Also what about if we encounter perofrmace with SRDF, is there any tool to analyze the performance issues other then the EMC's internal tool "symmerge" ?

What about prosphere, any thought on that?

Right... I don't know actually. All I know is that MS SQL has 2 sorts of disks: database and logs, but what kind of logs they are, I actually don't know ;-)

And about Oracle I know even less.

Do we have to deal with redo logs different as with archive logs?

When storage is assigned for a database server, right now the database administrators put the archive log on a RAID volume... When I am searching for the redo logs, I can find them on a RAID5 volume, with also the database on it.

I never paid attention to it. I was thinking our database admins knew what they are doing... ;-)

Exactly my thought , but with Oracle this might be different, I don't know, but it changes nothing to the fact that measuring your needs is always the best.

I think Analyzer is great. We have it running 24/7 and whenever we have problems I collect a week's worth of NARs (sometimes more) and I merge them into a single large NAR. Analyzer (set to advanced) will show everything I need, but also the AH (Analyzer Helper) is a great tool which helps me locating hot spots in an hour or so (depending on the machine I have this tool running on).

We also use Solarwinds for realtime monitoring and alerting purposes. On the host (Windows) perfmon is my preferred tool.

bartdonders wrote: Do we have to deal with redo logs different as with archive logs? When storage is assigned for a database server, right now the database administrators put the archive log on a RAID volume... When I am searching for the redo logs, I can find them on a RAID5 volume, with also the database on it. I never paid attention to it. I was thinking our database admins knew what they are doing... ;-)

we separate database, redo and archive ..not only from performance perspective but also recovery perspective. If your LUN that contains data files and redo logs becomes corrupt , you just lost transactions. You can restore from tape and then roll forward using archive logs, but they won't get you as close to the point of corruption as redo logs could if they were not destroyed.

Hi

I would like to hear if anyone can give some in depth information on performance about ISCSI and its use on the VNX.

We have the posibillity to get either 1G or 10 G ISCSI ports for the SP's on a VNX

My understanding is that everybody says that a 8G FC port can outperform a 10G ISCSI...And that in some cases 4 1G ports can out perform a 10G line...

My question is WHY ?

And from a performance side, what is an acceptable level of data transfer for the two port types....this will offcourse depend on the setup, but I'm looking for some kind of information on what to expect, and what not to expect in an ISCSI environment

Bart...

I think this white paper might be of help to you in your search for Oracle answers

http://www.emc.com/collateral/hardware/white-papers/h8242-deploying-oracle-vnx-wp.pdf

With Analyzer do the I/Os shown at the disk level include the parity calculation I/Os? For instance 1 write I/O coming from a host will result in 4 I/Os (RAID 5) on the physical disks. Should I expect to see those 4 I/Os when looking at the disk in Analyzer or will I see just 1 I/O?

We also run analyzer 24/7. Works well for when you hear about performance issues days after they occurred. Ideally we'll use all the data to do some system wide trending as well. Currently working on automating that process with a sql database and custom reports.

If you see the I/O at front end you will just see the actual number of I/O's but if you look at the back end disk I/O's the "yes" you will definetly see the I/O's with write penalty.

Exactly: it wouldn't be much of an analyzer product if it didn't show this! You can also see the influence of the cache filling up and the watermark flushing and about everything you always wanted to know.

Thanks for the document...

I will share this with our database administrators also.

Question: in the licensed version of Analyzer you can enable the collection of NARs for a maximum of 30 days. What do you guys use to monitor continuously? Do you manually disable / enable the log collecting or do you have this scripted?

I created a script for this, but I encountered vague problems. What I see is that disableing is not really an issue, but when I re-enable too quickly, it sometimes fails and I end up with not collecting any logs. I was thinking about building in some sort of "pause" to be sure the array is ready for the new command or a second option was to first disable it on all my arrays and then enable it on all arrays (instead of dis/en it per array and move on to the next array).

How do you do this?

Hi Jim, how was the weekend?

About iSCSI: unfortunately I don't have that much experience with it. The only time I encountered iSCSI I used it to copy data from a CX3 to another CX3 and the speed we got was about 30MBps, but then again, this was in the CX3 era, so about 5, 6 years ago or so. Back then colleagues of mine didn't get much more than 30MBps on a 1 Gb link either. So based on that I would say that FC work much more "clean" with less overhead so it can actually (almost) reach the theoretical maximum.

I wonder about the experiences of other people as well. Anyone?