What method to determine the performance between internal drives and NetApp luns on HPUX11iv3

[hpuxmagic]

We are comparing the disk performance between internal drives and NetApp luns on HPUX servers.
Could anybody please let me know what commands or tools to make such comparison.
I'd assume these tools/commands have to be running on HPUX.
OnTap 7.3.7
HPUX 11iv3@RX4640

Thanks!

[Duncan Edmonstone]

What sort of IO are you interested in?

Straight sequential IO? single/multiple streams?

Random IO? single/multiple streams?

Database IO (OLTP or DSS?)

Filesystem IO? Volume Manager IO? Raw IO?

something more specific?

There are various approaches and tools out there, but it pays to know where to start... different strokes for different folks!

[hpuxmagic]

I am looking for I/O on a file system which created on LVM. Basicaly which one would provide better performance, things like iops, or throughput, some data related to the performance.

thanks,

[Joseph P. Smith]

Here is a method that provides a framework for assessing proposed hardware changes.  This approach contributed to a hardware upgrade that satisfied user demands for several years, while cutting the vendor's proposed upgrade costs by some good percent.

I'm assuming that there is an application in here somewhere (database server, ERP, CAD, GIS, whatever).  You must understand your present situation before understanding the difference any change would make.  You have two bottom lines: money and actual performance. 

1. How much improvement in performance is needed to justify the cost of purchase, installation, and maintenance of the NetApp luns?  Is performance so bad that work can't be done, or would better performance improve some productivity measure enough to justify the cost?  Talk to the key users and find out what they need.  Try to quantify costs of continuing without the change in real $$. OTOH, are users pulling their hair out due to bad performance?

2. Before doing any modeling you must make sure that your present environment is optimal, i.e.. tunables are good, patches applied, database indexed well, etc. Etc.  You may not have to throw hardware and $$ at the problem!

3. How precise does your comparison need to be?  Can you configure two simple and *identical*  prototypes with the exception of storage?  If yes, then duplicate application config, user profiles, system CPU and memory loads, RAID levels, file system layouts, host-based mirrors, etc.  The only variable will be the disk/storage configurations.

4.  Now compare runtime for the specific work to be done.  Regardless of the storage system attributes on either old  or the NetApp system, this has to be your most significant metric and will be pretty precise.

5. If #4 cannot be done, then you do have to dig into the weeds and specify as many attributes of the existing environment as possible, and understand those attributes' contribution to overall runtime.  At that point you can then deduct from the runtime  the theoretical improvements that NetApp my claim to make using the new luns.  This is less precise.  Precision would increase with real expertise making the analysis.

5. Sometimes  you don't have to re-invent the wheel.  Does NetApp have a modeling tool that you can use?  Also, go to the application user group and find out how others have optimized performance across similar configurations.  Verify.    IMHO, if the answer has to be made in two weeks, then go with #5.

Pure numbers from NetApps sales rep, web page or catalogue never take into account the complexities of your run-time environment.  This is only my method to an answer, change and adapt as needed.

[hpuxmagic]

Joseph, Your messages very well said in high level on my questions. They help a lot.

The application running on the server is Informatica. One of file systems is lack of space, and we are planning on moving it to a NetApp lun. That will be only change. NetApp performance tool may only provide us the matics on it's own lun, not on the local internal drives the file system use to reside on. "glance" or iostat should be the ones to allow me to make comparison, but I am not sure what attributes I should look into, and what guidelines are to anayze these attributes.

Thanks!

[Joseph P. Smith · Accepted solution]

Thank-you for kind words.

You wrote:

>>NetApp performance tool may only provide us the matics on it's own lun, not on the local internal drives the file system use to reside on." 

Good - This gives you the chance to measure "apples to apples".   Buyer beware however; you'll be using NetApp's marketing tool and performance statistics.   It may be optimized for its own devices/purposes (sell, sell, sell).  This is the grain of salt with which you may want to take its results.

You can take the identical metrics that the  NA tool uses (queue length, seek time, response, etc.) and acquire the same data from the system with glance.

My preference is glance, specifically Motif glance  ("gpm &").  You need an X server running on your desktop.  The motif check box helps you easily select metrics, etc.    Drill down to "Disk" -> "Reports" -> "IO by Disk" -> "Configure" -> "Choose Metrics".

That said, character based "glance" is good too. Other readers may differ on the tool & display of course, and have better/more useful suggestions.  Different HP tools/commands may use same metrics, but display different results, depending on the source of the data (hardware config, system counters used).

Given those cautions, when finished you can compare the two hardware configs.  IMHO - until/unless there's a clear understanding of the degree to which the disk/io subsystem contributes to overall job/task "speed"  ("How fast does this job run?"), it will be hard predict the effects when the NetApp disk is implemented on the production environment.

Sorry to be so generic, but that's my world ... "Joe Smith" ;-)