Re: oracle benchmarks on VMS

Hein van den Heuvel wrote:
>
> Tim Smith wrote:
>
>

>>I see Oracle publishes some benchmarks at
>>http://www.oracle.com/apps_benchmark/, but notably there is nothing
>>for VMS - is that because Oracle writes to the filesystem, not direct
>>to device file files directly i.e. VMS filesystem is a lot slower than
>>raw devices?

> The VMS Filesystem is actually an advantage! The VMS filesystem does NOT
> buffer data.
> On Unix systems the OS tends to waste time and memory buffering Oracle
> data pages which are better managerd by Oracle in its buffer pool (SGA).
> On many Unix implementation, for ultimate Oracle perfromance once has to
> deal with hard-to-manage 'Raw Devices' to avoid said buffering.
> On VMS you have the comfort of a file system for Alloaction, Naming and
> backups yet the speed of a raw device. On HP Tru64 Unix Oracle can (and
> will) use the DIRECT IO feature to get the same effect on single systems
> as well as in clusters.
>

On most UNIX's you now have direct I/O which avoids the buffer cache. Solaris UFS+ supports direct I/O, VxFS does, IBM JFS does etc.

And the buffer cache isn't always a disadvantage since it can act as a cache for reads but it doesn't help writes and it does consume VM resources.

Because most UNIX filesystems support DIRECT I/O and because differential between RAW and COOKED is now negligable very few people use RAW devices anymore.

With COOKED obviously giving you the advantages of log based filesystems, snapshots, growing filesystems, shrinking filesystems (VxFS) etc but with the performance of RAW.

Regards
Andrew Harrison Received on Tue Mar 04 2003 - 05:35:36 CST