Fixed: OpLocks caused the corruptions/slowness (Was: How Samba let us down)
seniorr at aracnet.com
Sat Nov 23 00:50:01 GMT 2002
>>>>> "David" == David W Chapman <dwcjr at inethouston.net> writes:
jra> Clients commonly ignore oplock breaks because of network problems
jra> (borderline hubs etc.). Many people are suffering from network
jra> hardware that performs adequately in light use situations and
jra> fails under heavy load. I myself have ended up junking hubs with
jra> this problem.
Russell> I *still* don't understand how flaky hardware could be the
Russell> problem. TCP connections are supposed to be reliable. If
Russell> flaky hardware is eating packets, then surely the sender,
Russell> failing to get a timely ACK will resend? I can understand a
Russell> flaky client getting the break and not responding with the
Russell> appropriate action, but I don't understand how a hardware
Russell> level problem can break TCP communication, except in the
Russell> obvious and persistent way of not having a connection at all.
Russell> Please someone, wump me with a clue stick.
David> Duplex problems are a simple example like setting full-duplex
David> on a half-duplex switch might.
David> A bad cable or NIC could cause intermitten problem as well.
David> Yes packets would get retransmitted, but who's to say the
David> retransmitted packets won't get dropped as well?
Uh, RFC793, section 2.6 maybe? I think people would notice a total
network failure, so I am excluding that possibility.
Russell Senior ``I've seen every kind of critter God ever made,
seniorr at aracnet.com and I ain't never seen a meaner, lower, more
stinkin' yellow hypocrite than you!''
-- Burl Ives as Rufus Hennessy
More information about the samba-technical