On 06/10/2011 09:23 AM, Boaz Harrosh wrote: > > I disagree. Please make that same exact patch for the server you are using! > > Leave the client alone. Don't even consider getting use to this, sticking > broken stuff on client scripts. If anywhere it should be in Server configuration > files > > Boaz > BTW the algorithm at the *server* side need not be dynamic and can be very simple. Just a simple rule based: 1. If the file is new/zero-size just give out a small layout say 1-2 stripe_units 2. If the file is or becoming bigger give out some BIG maximal optimum IO size. A size that anything bigger will not increase performance. 3. If the file is opened by a second client, recall the first big layout and give out a one-to-few strips layout for shared access. You will find that this shared file case is very rare. There are not many application that share the same file. If they do they usually use range locks. Query your lock manager if the client has a lock on this range of the file, give out the full locked range as a layout. If that is not the proper hint from the application, then what is? It is a much better hint then the nfs-client can ever guess. 4. Make it simple as hell.... Just my $0.017 Boaz -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
