[Date Prev][Date Next] [Thread Prev][Thread Next]
[Thread Index]
[Date Index]
[Author Index]
Re: RHEL 4 improvement in use of swap? (compared to RHEL 3)
- From: Tom Sightler <ttsig tuxyturvy com>
- To: "Red Hat Enterprise Linux 4 (Nahant) Discussion List" <nahant-list redhat com>
- Subject: Re: RHEL 4 improvement in use of swap? (compared to RHEL 3)
- Date: Fri, 28 Oct 2005 16:37:58 -0400
On Fri, 2005-10-28 at 15:22 -0400, Brian Long wrote:
> FYI, I was wrong about pagecache. The old values were 1 15 100 and the
> new values are 1 15 30. This means pagecache will only occupy 30% of
> your total RAM.
And, on some of our server, we find even this to be far to much. We
have several servers running http/java applications serving hundreds of
concurrent users and they were swapping like crazy during peak loads.
We added the following to our RHEL3 systems and things improved
dramatically:
# Make system more likely to release pagecache rather than use swap
# Without this system can go to 100% swap usage.
vm.pagecache = 1 1 2
Without this RHEL3 was far too unlikely to release pagecache. These
systems have 8GB of RAM and 2% is still 160MB of RAM for the pagecache.
This is more than enough for many applications servers that don't use
much IO.
With the default settings of 30% what we found on these systems is that,
under periods of low load, and especially when running backups, the
system would have enough free memory and allocate it's 30% to the
pagecache. This is 2.4GB on an 8GB machine and that's fine when under
low load, however, as the number of users increased, the system started
swapping rather than releasing pagecache even though almost no IO was
actually occurring (except for the swap of course). We found that
significantly limiting the pagecache was a complete win on these
systems.
Later,
Tom
[Date Prev][Date Next] [Thread Prev][Thread Next]
[Thread Index]
[Date Index]
[Author Index]