From: Craig_Everhart@transarc.com
Date: 02/05/99-12:52:43 PM Z
Message-Id: <QqinrveSMVcC0bn8o0@transarc.com> Date: Fri, 5 Feb 1999 13:52:43 -0500 (EST) From: Craig_Everhart@transarc.com Subject: Re: NFSv4 and Caching Fascinating message--thanks for spelling it all out. Let me amplify the payoff for callbacks by observing that lock-management protocols can benefit greatly from the same callback mechanism. Without a callback mechanism: (a) clients must return locks to the server as soon as an application unlocks (b) clients must poll in order to obtain locks under contention With a functioning callback mechanism, the server can notify a client if the lock that it holds is desired by another client; until then, the first client can cache the lock. (Lock caching, in principle good anywhere, is particularly attractive for files that are actually being used from only one client.) In addition, callback messages can convey locks to clients at the time when they are available, instead of clients having to poll for them. One aspect of NT semantics is basically mandatory locking, so in that context, file locking is as prevalent as reading and writing. Thus, for these clients, caching of locks is as valuable as caching of data. It's an excellent challenge as to how callbacks could fit into NFSv4 to facilitate low-overhead caching. Craig
This archive was generated by hypermail 2.1.2 : 03/04/05-01:46:38 AM Z CST