From: Brent Callaghan (brent@jurassic)
Date: 03/22/97-11:13:39 PM Z
Date: Sat, 22 Mar 1997 21:13:39 -0800 (PST) From: Brent Callaghan <brent@jurassic> Subject: Re: FH -> filename Message-ID: <Roam.1.1.859094019.13894.brent@jurassic> > Carl Beame writes: > I have to disagree that it is too expensive. If it took 10 seconds per > file and put a 10% load on the server it would be worth while to > implement so that major problems like outstanding locks can be cleared > and people can get back to work. I have to agree with Mike Eisler here, it can take *much* longer than 10 seconds to do the reverse mapping. Unix (and perhaps most OS's) are generally well set up to map filenames to inode numbers (or disk addresses) very quickly. However the reverse is much harder. Unix, for instance, can do the reverse mapping only via exhaustive search. On a large filesystem (aren't they all large these days ?) this can take an inordinate amount of time. The best policy is to avoid the need to do this. For instance, our client in Solaris logs some errors by dumping out the filehandle in hex. This is hardly user-friendly - far better to display a pathname that means something to the user and usually the client's pathname is more meaningful to the end-user. Brent
This archive was generated by hypermail 2.1.2 : 03/04/05-01:45:36 AM Z CST