How to measure performance difference.

[kumasudh]

Hi,

I am having a rx3600 itenium server having a 73GB hard disk. I want to replace this disk with 146GB hard disk. But before that i want to calculate the performance difference which could been seen my my application.

P/N Description
AD333A HP Integrity 146GB 10k SAS Drive
AD379A HP Integrity 72GB 15k SAS Drive
As clearly stating that the 146GB will have poor IO performance compared to 73GB as 146GB supports 10k rpm adn 73 GB is 15k rpm.

could anybody help me figure out how to benchmark the performance of my application with 146GB hard disk and with 73GB hard disk.

Thanks,
Best Regards
Sudhir

[Steven E. Protter]

Shalom,

Best way to measure performance is to create a realistic test and conduct it with both configurations.

If the server is for example an oracle server, then a realistic test should be conducted running the exact same oracle transactions in both scenarios.

On a most basic level, I'd take the same 20 GB file, and copy it to the location, with a script measuring stop/start time. I'd run this test more than once to insure that the file is buffered.

What you have not provided in your post is how you came to the conclusion performance is poor. That would be helpful in designing a more specific performance test.

There are performance measurement tools on the market. I even have a free script set that you can run while conducting performance benchmark tests. http://www.hpux.ws/?p=6

SEP

[kumasudh]

Hi Steven,

Thanks a lot for the reply. My assumption behind the conclusion for saying that 146GB disk will give poor IO performance ,was that my applicaition does lots of disk access in terms of database update which is sitting on the same disk. And having said that the rpm is less for 146GB disk(10k)compared to 73GB (15k) so my application will definetly feel a slow down in terms of disk usage.

Best Regards
Sudhir

[Torsten.]

Only a realistic test will give you this information. I doubt if you can calculate anything.

This server can have different controllers that connect the disks, so read and write cache of the controllers comes into play more or less in addition of striped logical drive configurations. For this reason the disk access may be fast even with "slow" disks.

[Emil Velez]

Alot depends also on buffer cache.. How much memory does your system have and how much of it is used by buffer cache. If your access pattern references blocks and files already in buffer cache then the disk speed is not a issue.

This is even for reads and writes.

Emil

[Bill Hassell]

For any computer, a bunch of small disks spinning at a faster RPM will be faster than one large disk. The reason is simple: there is only one path to the disk and the disk can only perform one read or write at a time. On the other hand, you have only one disk so you will NOT have poor performance between the 10K and 15k disks. A continuous series of reads (not common in Oracle) will be slightly slower (not poor).

As with all benchmarks, you must create some sort of sql query that takes a while 10-50 seconds to complete. Because HP-UX uses a buffer cache, the first run will be a lot slower and then subsequent runs will be faster.

But as with all database products, you can configure the database engine to trade off disk I/O for large amounts of RAM. This will be far more important than one particular disk. Large amounts of RAM for Oracle is something like 8 GB or more. Your DBA will have to configure Oracle to use the extra memory.