We are using a shared file across a distributed network for
process scheduling. It appears that after the file is locked,
read, modified, written to, fsynced and then unlocked, there
is a small delay of 1second during which any read occuring
from another machine will get the old file before modification.
This sort of defeats the purpose of the file lock and I thought
that 'fsync'ing would flush any NFS caches before it returned.
Does anyone know why the cache would be persisting for that
long? We are doing this under Irix5.3 on Indys on a standard
NFS filesystem (No CacheFS and no Automounting).
Is there a workaround for this (apart from putting an inherent
delay into the filesharing protocol?)
Lecturer in Computer Animation,
Australian Centre for the Arts and Technology (ACAT), Institute of the
Australian National University, Canberra, ACT, Australia.