Maximize thoughput on Oracle DB and HP3640 snf EVa5000

[Tim Jacobs_3]

Can I maximize the I/O between Oracle and the EVA by adjusting the size of requested data. The database size is 1 peta byte and We need to maximize the throughput.

I remember for the SUN 15k we used "ndd" and adjusted something. Is this needed to optimize the data thoughput in the HP equipment. We are running new software Oracle 10g, Hp 11.23 and not using service guard.

[B. Hulst]

You can adjust the oracle blocksize to match the n*(size used on the diskarray).

Of course not so easy to do with a running system. Only with a new creation of a database.

[Tim Jacobs_3]

I am using the MAX sizae for Oracle at 32K, and with a multi_block_read_count of 32 that means we read 1 meg at a time. I am trying to find the optimal amounts to set Oracle to maximize the throughput. I have no idea what the FiberChannel uses and what it would like for maximizing throughput.

I need someone that really knows the storage at a lower level.

[Uwe Zessin]

A Fibre Channel frame can be up to 2Kbytes big and, if I recall correctly, FC can transfer up to 16 MBytes in one request, but not every system can do such large IOs.

[B. Hulst]

You could just try reading with dd data from the diskarray and try using different block sizes and time it.

for i in 512 1024 4096 8192 16384 32768
do
times dd if=/dev/array of=/dev/null bs=${i}
done

It will give you an indication of the optimal raw data speed when reading from the diskarray.

And take a file which is larger then could possibly fit in the diskarray's cache, so it has to come somehow from disk.

[Tim Jacobs_3]

Yes, Not all systems can do these large transfers. That is why I came to HP, to see if this Storage Forum would be able to answer this. I can use the dd like one person responded and keep the times.

Can you change the read block size when using FiberChannel? or is that a Ethernet only thing.

What we have are several HP EVA5000's and that is what we are trying to maximize.

The actual system is a write-once read many system with over 1 million rows inserted a second. That is why we need this optimal.