From: Xuan Baldauf (xuan--reiserfs@baldauf.org)
Date: 02/04/01-04:22:45 PM Z
Message-ID: <3A7DD634.B7B44EF1@baldauf.org> Date: Sun, 04 Feb 2001 23:22:45 +0100 From: Xuan Baldauf <xuan--reiserfs@baldauf.org> Subject: Re: [reiserfs-list] nfs standards guys still screwing up in definition of cookies, it's still a thinly disguised hardcoded byte offset Hans Reiser wrote: > [...] > It is very doable, and essential to allowing decent optimizations of > directories. We are going to delay implementing filename compression in > directories until disributed reiserfs exists primarily because of NFS being a > problem, and even then we will have to make it an option so that those who need > NFS can use it. > > Hans Maybe a directory format plugin would be desirable. People needing stable NFS use the hash format (with the various hash algorithms). People needing fast access to their thousands of subdirectories use alternative formats (e.g. stem compression, everything you like) at the cost of being not fully NFSv2 compatible (and having a question mark at NFS in general). The cases where both needs (NFS and large directories) collide are much more rare than the cases where the current hash format and large directories do collide. But be cautious, if you implement directory format plugins, plugins for other things (like stat data emulator), etc., we can end up in SuperFS. *grin* Xuân. :-) P.S.: What will be special with distributed reiserfs? Could a modified|better samba|smbfs do equal work?
This archive was generated by hypermail 2.1.2 : 03/04/05-01:48:29 AM Z CST