1 |
On Thursday 22 April 2010 18:24:08 Stefan G. Weichinger wrote: |
2 |
> Am 22.04.2010 17:50, schrieb Alan McKinnon: |
3 |
> >> Shouldn't the kernel *swap* then ? |
4 |
> > |
5 |
> > No, the OOM killer kicks in when the kernel has no more virtual memory, |
6 |
> > including swap. Either way, more RAM is the answer. Or fidn the app with |
7 |
> > the memory leak if you are unlucky enough to have one of those running |
8 |
> > around. |
9 |
> |
10 |
> The added swapfile with one GB won't help here for a start? |
11 |
|
12 |
It will certainly help. If your core problem is simply not enough RAM, then 1G |
13 |
more might be all you need. You'd have to run checks and do some monitoring to |
14 |
see if performance is affected. |
15 |
|
16 |
I haven't followed the full thread so I don't know what you are running; and |
17 |
some daemons perform really badly if they have to touch swap. Apache for |
18 |
example, a busy MTA for another - disks are thousands of times slower than |
19 |
RAM, so if a webserver has to swap memory back in from disk, it almost |
20 |
instantly brings the server to a grinding halt. |
21 |
|
22 |
On my web and mail servers I have no swap at all, they do have lots and lots |
23 |
of RAM; my Sybase database servers have enormous amounts of swap. Each server |
24 |
has been profiled so it is set up to be as close to ideal as I can determine. |
25 |
|
26 |
-- |
27 |
alan dot mckinnon at gmail dot com |