<snip>
This is generally why your group size and OS
page size should be the same (yes, I know...unless you have some kind of
giant records, or dynamic files, or any of the other "YaBut's").

> So...this school of thought says that modulo 1 is a good thing, and
> that it forces files into memory.  The question is -- is the school of
> thought speculation, or is it true?

Whether to size files to OS page sizes or size the file to whatever
guide says is one of those subjects that we should invest some solid
benchmarking in as comments for and against seem to come up from time
to time.  I have a vague recollection Stephen O'Neill at IBM has
posted on this but I'll need more coffee to say for sure.

My limited understanding is...

If the file group is bigger than the page size then the OS will need
multiple reads to read it.  However, Linux (and I presume any modern
OS release) do predictive read-ahead and if you have a SAN, it should
also be doing the same thing.  Factor in the buffering also going on
in the OS and SAN and the answer is less clear.  While multiple read
requests may go out for the file group, if it is in an OS or SAN
buffer it will be coming back fast.  So on a properly spec'ed machine,
sizing the files according to guide and leaving the OS to do its job
is the best course of action?

Correction or comments welcome.

Regards,

Adrian
-------
u2-users mailing list
[email protected]
To unsubscribe please visit http://listserver.u2ug.org/

Reply via email to