Performance on rp7420

[David_711]

Hi,
I have a rp7420 server with 2 virtual partition, two cells with 8 and 20GB. My principal virtual partition have 16GB and 8 processors.
I want to know the real performance transfer rate on my system.
I used the utility hp_ltt, this have the opcion sysperf to know the read rate and write rate to the disk subsistems. This utility show me rates between 32mbps to 60mbps
when i choise internal file systems how /opt or /tmp.
My volumen manager is veritas 3.5mp3 on hpux 11.11.
I don't know why i obtain this poor perfomance internally.
This tool is ok for this use?
Do i have a problem with my systems?
Every one know how i can do to know the real internal transfer rate on mbps and what is the real rate for a rp7420?
The check is necesary because i saw my system with my data base slow.
I do the same checks with my file systems on my san with a hitachi a the rate are simillar.
Please help.
Thanks

[Steven E. Protter]

Shalom,

You will need to run some large file ftp tests to get realistic speed information.

Other things:
http://www.hpux.ws/system.perf.sh

Performance document: I saved this as web page because HP keeps making it disappear. Nothing personal:

See attachment:

SEP

[Torsten.]

The hp_ltt you mentioned is the library and tape tools (LTT) set, intended for use with tapes. I'm not sure about the precision of the results (from the user guide: "The backup pre-test measures the read performance of the disk file system (simulating a backup)..." ).

Anyway, have a look at the servers quickspecs for the sustained bandwidth.

http://h18000.www1.hp.com/products/quickspecs/11895_na/11895_na.HTML

[Bill Hassell]

Measuring performance on a large scale SAN is always difficult. If you read a 500 meg file one time, it might transfer at 50 Mbytes/sec. But read the file once more and the transfer rate will be 200 to 300 Mbytes/sec for the same file. The reason is that most of the file is in two caches, one in the Hitachi and one in HP-UX. Generally speaking, a slow database is almost always due to poor design such as lack of indexes, unbalanced indexes, very small SGA, etc. You need Oracle's statspak to identify the delays and HP's Glance to show you the true performance on your system.

[Alzhy]

If your systems use 2GB Fibre channel cards and Gigabit ethernet cards AND are doing filesystem based transfers over your GigE connections -- THEN your measured throughput of between 32 to 60 MB/s is just about right.

Look no more.