[ Date Index ] [ Thread Index ] [ <= Previous by date / thread ] [ Next by date / thread => ]
On Tue, 21 Jun 2005 09:39:11 +0100 John Daragon wrote: > I thought I'd just chip in here because a couple of preconceptions are > getting in the way of finding out what's happening to your system : Cheers. The more the merrier! > 1) nice. nice is a mechanism for decreasing task priority. This > effectively decreases the probability that a task (if runnable) will be > scheduled on return from interrupt. It's of marginal use in the control > of tasks which perform much I/O. In this case, with low CPU usage on the > machine, you can reduce the priority of this task as much as you like, > and it'll still get run when it's not in iowait. Yeah, got that. I think this had been established, but thanks anyway. (renice can also raise the scheduling priority of a task if run as root.) > 2) Paging. If you have no swap space you will not be performing any page > out operations. If you perform no page outs, then you will be performing > no page ins, either (with the exception of loading executable code). So > the I/O you are observing is not paging. > > Of course, given that you can't page out, the amount of space you have > for block buffers may be limited. What you *may* be observing is a > (relative) failure of the block cache to perform efficiently because it > has limited space to work with. The block/paging cache scheme varies > from kernel to kernel, and I haven't looked at the Linux one for a > while, so I'm disinclined to offer a concrete mechanism. > > jd Cheers John. One quick observation though: whenever this problem occurs, kswapd is *always* put quite near the top of the, erm, "top" output, so I would guess that it's a currently running daemon rather than a dormant one. Since I have no swap, is it possible that kswapd is actually doing something anyway, which is causing this problem? (With any luck I will be re-installing the other harddrive soon so I can have a swap partition again... yay!) Grant. -- Artificial intelligence is no match for nuratal stidutipy. -- The Mailing List for the Devon & Cornwall LUG Mail majordomo@xxxxxxxxxxxxx with "unsubscribe list" in the message body to unsubscribe. FAQ: www.dcglug.org.uk/linux_adm/list-faq.html