Keith G. Murphy
2003-Feb-27 17:41 UTC
[Samba] Could Linux load average problem be related to smbfs?
This is more a "has anybody seen this?" question than anything else. I'm using smbfs version 2.2.3a-12 on 2.4.18 kernel, Debian 3.0. The other day, I had a problem where df got hung in D state because of smbfs mounting a share, then the PC exposing the share rebooting. (This has happened with smbfs across several versions). umount gave "device is busy" errors, and I couldn't kill the df processes; I used 'umount -l' to work around the problem. The weird thing is, I noticed less than a day later that the system was getting about an 8 load average, with very little actually running. All those df processes were still around of course, still in D state. Thanks for any comments.
%%jrrs
2003-Feb-27 19:34 UTC
[Samba] Could Linux load average problem be related to smbfs?
On Thu, 27 Feb 2003, Keith G. Murphy wrote: [ getting about an 8 load average, with very little actually running. ] this might be wholly inapplicable, but i once had a situation similar to that, where my reported load was much much greater than my perception of what the system was actually *doing*. i had enabled the diskd cache whatnot method of squid, rather than the normal ufs method. either the squid process itself or one of its child processes was polling something ( i don't believe it was the physical disk, but i don't fancy being quoted on that ) once every second or so. the poll was only a blip, but it was enough to keep the load high. so, perhaps if those processes were strobing/polling something, it wouldn't make the system run as busily as the load was telling you? again, that might not be terribly applicable. ? jared.