[5081] in Athena Bugs
afs panic
daemon@ATHENA.MIT.EDU (don@ATHENA.MIT.EDU)
Tue Jun 5 21:15:07 1990
From: don@ATHENA.MIT.EDU
Date: Tue, 5 Jun 90 21:14:41 -0400
To: bugs@ATHENA.MIT.EDU
Cc: probe@ATHENA.MIT.EDU, roden@ATHENA.MIT.EDU
richard suggested that the getdcache panic comes from the kernel
running out of file-cache entries, which is distinct from the cache filling up.
he suggested that i add a "-files nnn" option to /etf/rc's afsd startup,
because afsd currently guesses that it'll need 1 file-entry for every
8 cache-blocks, and because this option overrides afsd's guess.
this worked; i have a cache size of 500 on drgonzo, so i tried values
of 50, 100, 128, 150, 250, and 500. the default would be around 64, which
is hypothesized to be the source of failure. sure enough, 50 fails
egregiously, 100 fails about half the time, and the higher numbers
cause no trouble so far. i'm currently running with 128, so as not to
make the maximum cache-file size too small.
probably, afsd wants to force a minimum value on the number of file-entries.
since the default cache-size is 2000 (?), and since 1/8th of 2000 is 250,
we shouldn't expect trouble in the field with 7.0f's afsd code.
-don
