Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Mailing Lists -> Oracle-L -> Re: OT: question about sizing swap for solaris

Re: OT: question about sizing swap for solaris

From: <dantow_at_singingsql.com>
Date: Tue, 4 May 2004 10:24:46 -0500
Message-ID: <1083684286.4097b5bea39f9@www.singingsql.com>


Responses inline...

Quoting zhu chao <chao_ping_at_vip.163.com>:

...
>
> To Dan Tow, your words said that if swap size is small, application will
> hit "out of memory", while that whitepaper on sun.com says, more swap space
> than needed can actually cause hard-pageout to occur more easily. There is
> some difference between your opinion and the paper.

Well, I haven't read the still-unreferenced paper, but absolutely there is a difference between my opinion and what the paper apparently says - to be clearer, in my opinion: low swap size "prevents" page faults (by triggerign errors, instead) only in the same sense that running out of gas "prevents" bad gas mileage!

> And you said , from experience, you saw some application with a lot of
> processes swapped out to disk, without performance impact on end user, is
> that an oracle database server? I think that will badly impact on system
> performance, according to my experience(Of course my experience is much less
> than that of yours).

My statement applies to database servers and non-database servers, alike. Where I have specifically, quantitatively measured the true end-user costs of paging and swapping, I have never seen them to be high, except on systems that were victims of operating-systems swapping-algorithm bugs, which have long-since been fixed. I *have* heard plenty of *anecdotes* about swapping and paging being big problems, but I have never seen these stories backed up with hard data. I certainly believe it's *possible*, though - you just need to buy less memory than you need, in spite of how cheap memory is. I have gathered data on a different subset of systems than you have, however, so it is entirely possible you have seen one of those hypothetical systems with insufficient memory.

>
> Regards
> Zhu Chao.



Please see the official ORACLE-L FAQ: http://www.orafaq.com

To unsubscribe send email to: oracle-l-request_at_freelists.org put 'unsubscribe' in the subject line.
--
Archives are at http://www.freelists.org/archives/oracle-l/
FAQ is at http://www.freelists.org/help/fom-serve/cache/1.html
-----------------------------------------------------------------
Received on Tue May 04 2004 - 10:23:40 CDT

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US