-----BEGIN PGP SIGNED MESSAGE-----

In article <199505111641.JAA26631@Pescadero.Stanford.EDU>,
Jonathan Stone <jonathan@DSG.Stanford.EDU> wrote:
>I used to see this, and I found it was cured by turning off read-
>and write-clustering -- either via sysctl in rc.local, or (more permanently,
>modulo sup updates) by hacking the kernel source. I don't recall
>seeing it since, either on a machine *with* swap space, or  one without.
>(Both have 32 Mbytes of memory.)

Now that I've had a chance to watch some kernels with clustering turned
off over time, I've noticed that though many of my problems have decreased
dramatically in frequency, they haven't disappeared.  

...my conclusion: that the read/write clustering isn't the source of
the problem but rather tickles a bug elsewhere.  I keep looking out of
the corner of my eye at VM, but am intimidated at going deep into it
for now (BTW, where can I get docs on 4.4 vm?  Ted mentioned to me that
the demon book is obselete on vm).

BTW, I've also noticed that the memory in the machine makes a big difference
in the frequency of all the problems (inetd through crashing under load).
My machine with 24 MB crashes a whole lot less often than the 4 with
16MB (all 2100s... in a couple of days I'll have some 5000s to play with).
- --
Todd Masco     | "If we don't make utter fools of ourselves from time to time,
cactus@hks.net |  we grow smug - that is, we do not grow at all." - T. Peters
     <a href="http://www.hks.net/~cactus/cactus.html">Cactus' Homepage</a>

-----BEGIN PGP SIGNATURE-----
Version: 2.6.2

iQCVAwUBL7JwBhNhgovrPB7dAQGgLwQA2NaoHblf1+QYHOyCcAUpMRw16NlZfg93
pfmMHN1QtCVz6XczepWbNkdvdu9xFuWY1cLYxlCb98GrSRpkcFhrR37L/9zESVry
8gNbZkf9TgOnRkOFl3nCd4JLPE2drC3JoGcnzIrRDFWHy/kkNoxHIBZ3W3VKoeLb
KKbr9ca3/Mo=
=rxXB
-----END PGP SIGNATURE-----