On Thu, 16 Dec 2004, John Goerzen wrote:

On Thu, Dec 16, 2004 at 03:54:46PM -0600, Jon Nelson wrote:

I encountered something very strange last night similar to other reports of starvation (at least different reports by John Goerzen and others, back to June and July).

Interesting you should mention that, as I've seen a few mild cases myself.

Let's compare notes...

I've got 3x 80G IDE drives in a raid5 array (150G or so usuable).
On top of that I've got LVM.

No RAID going on here, but I do use LVM everywhere I possibly can.

[ snip ]

Even an ls in a directory with < 20 entries would take 20+ minutes to
complete.  This is on a Duron 750 with 1 GB of RAM.  Anything involving

Very similar to what I saw, but much more extreme.

and the system was totally usable - it would appear that writing (both

I noticed my problems during runs of updatedb (for locate). That process basically reads a bunch of directories, probably stats all the files, but never actually opens the files. (I think it runs find.)

I wonder if this has something to do with directories or inodes.

Hmm. That /is/ interesting. One of the things I copied over was my squid cache. It's not huge or anything (1.8G and ~100K files), but it /was/ part of the same "process" (cp -a) that copied over .flac files and some other very large files.


I wonder if the kernel attaches to processes some list of "files it has done something with" and the longer that list the more expensive life gets.

What's odd is how /one/ process was seemingly able to exclude others from accesing the disk, but only while running (suspended things went back to normal).

(BTW, how's Boa going along these days?  I think it was probably about 6
years ago I first contacted you about it.)

It's going very well. For about 3 years now I've been telling coworkers that 1.0 is "just around the corner". yeah, right. AFAIK, Boa is the oldest web server in use (predates Apache, you know) and still hasn't hit 1.0.


--
Jon Nelson <[EMAIL PROTECTED]>
_______________________________________________
Jfs-discussion mailing list
[EMAIL PROTECTED]
http://www-124.ibm.com/developerworks/oss/mailman/listinfo/jfs-discussion

Reply via email to