Quote:
ISTR the old problem with NFS write speed was that the server was required by spec to only report a completed write once it had actually been written to disk - not any disk cache. That was back on NFSv2, anyway...



On Linux, this is now controllable via the "sync" and "async" NFS mount options. Use "sync" for slow, guaranteed-to-be-written access; use "async" for fast report-back-once-you've-got-it-cached accesses.