Re: oom-killer vs system not reachable
On Sun, 19 Dec 2004, Robert Ian Smit wrote:
> not work with it. There is so much mail that IMAP would be way to
> slow. Mutt is very fast once the mailbox is open. Reading and
Ha! Trash like uw-imap would be slow. Try it with Cyrus IMAPd 2.1, it is
packaged and if you follow the README.Debian.simpleinstall, you can have it
up an running in 10 minutes. I mean it. But you will *need* a very good
IMAP client (mutt is a piss-poor IMAP client) to test the performance.
It cannot be faster than the perfect client (i.e. one that also indexes
folders) accessing the filesystem. But if your client does not create
indexes, it just might be faster to talk to Cyrus over IMAP than go directly
to the filesystem. It also depends whether you just read the last message,
(in which case nothing can beat direct access to maildir), or do searches
often.
You'll have to use something like mutt to upload the messages to Cyrus (by
selecting all messages and doing a save to folder into the IMAP server).
Since this is an one-time thing, mutt will do just fine.
> > And make sure you use reiserfs, or xfs, or 2.6.9 ext3 with all the
> > hash tree options enabled if you're going to get 20000+ messages
> > in a single directory.
>
> I'll look into it.
For Cyrus or Maildir you will *need* it. ext3 without all the hashes will
be *horrible*.
> I will add another swap partition to see if it changes anything.
> Mutt is bound by a filesize limit of 2GB. Current memory
> configuration is 1 GB RAM and 1 GB swap. Is there a limit to how
> much swap I can/should add to the system?
Not really. It will just not be all mapped to RAM at once. I have some 4GB
RAM machines that have 4GB of total swap space (due to sheer paranoia, the
machines rarely swap at all and everything in them is swap-friendly, i.e.
they keep working in a relatively small set of pages most of the time), and
it works faillessly.
> The points concerning the mbox file are taken (in fact I knew as
> much). Do you think that Linux might stop to respond or get stuck
> for a while as a result of the big file.
If it is swapping? Depending on the kernel and swap device, yes, the
machine will slow down to a crawl.
> In other words would the symptoms of an oom-killer event and the
> resource shortage before it be noticable for a longer period? Or
> would it just be a quick kill of a process and then everyting back
> to normal?
It depends on what the OOM killer choses to kill, doesn't it? If it kills
mutt, it is an one-time thing.
But if your swap performance is poor, the machine will trash quite a bit
before the swap fills and the OOM killer is run (and adding more swap in
that scenario actually *hurts* more).
--
"One disk to rule them all, One disk to find them. One disk to bring
them all and in the darkness grind them. In the Land of Redmond
where the shadows lie." -- The Silicon Valley Tarot
Henrique Holschuh
Reply to: