[Date Prev][Date Next] [Thread Prev][Thread Next]
[Thread Index]
[Date Index]
[Author Index]
Re: "Out of Memory: Killed process" errors on server running Oracle or VMware
- From: Tom Sightler <ttsig tuxyturvy com>
- To: "Discussion of Red Hat Enterprise Linux 3 (Taroon)" <taroon-list redhat com>
- Subject: Re: "Out of Memory: Killed process" errors on server running Oracle or VMware
- Date: Wed, 25 Jul 2007 22:42:38 -0400
On Wed, 2007-07-25 at 19:45 -0600, Eric Sisler wrote:
> Well yes, I was talking about the potential for overload from just the
> VM standpoint, not VMs + the OS. My test server also isn't anything I
> would configure for the real-world. I intentionally increased the RAM
> on the VMs to try and trigger an oom-kill or system halt. Normally this
> particular server would have VMs configured for ~5Gb.
OK, cool, I guess I missed the part about this being a test only setup.
I was thinking it was just another server you were having the problem on
and I was just wanted to point out that it wasn't configured in a way
that would be ideal. Obviously for a test that's not really an issue.
> On a production server I wouldn't have any reason to set VMware to
> "allow most...". Assuming a production server with a reasonable number
> of VMs for the available RAM, are there any dangers to using the
> lower_zone_protection option you can think of or have experienced?
None that I'm aware of. It's a pretty common setting, just do a search
for it on Google and you'll find it mentioned regularly as a solution
for the OOM problem, although, like anything on the Internet, some of
it's wrong. The only real thing I can think of is that if you set it
too high you'll waste a good bit of memory that will end up not being
used for anything. For example, if you set it to 500, that will
"protect" about 500MB, but it's possible that 450MB of that would never
be used for anything which is a waste of ~5% of your memory on a 8GB
machine.
It would probably take some playing to find the ideal setting that will
keep your system stable, and waste the least amount of memory. BTW,
newer 2.6 kernels, like the one in RHEL5, seem to use a new setting,
lowmem_reserve_ratio, which appears to default to protecting at least
some memory in every zone by default.
If you decide to try hugemem or 64-bit I'd be interested in those
results. That being said, if your over-committing memory by 3-4GB on an
8GB machine then it's likely you may still need to protect some memory
because of VMware locking the memory it allocates so that it can't be
swapped.
Later,
Tom
[Date Prev][Date Next] [Thread Prev][Thread Next]
[Thread Index]
[Date Index]
[Author Index]