[314] in Athena_Backup_System
Jonathon Weiss: Re: Minutes of 19 Sept Meeting
daemon@ATHENA.MIT.EDU (ted@MIT.EDU)
Thu Sep 26 11:22:30 1996
From: ted@MIT.EDU
To: athena-backup@MIT.EDU
Date: Thu, 26 Sep 1996 11:21:56 EDT
It seems I didn't think of all the different ways we might want to
make up dumpsets at our last meeting. Is it reasonable to allow
pattern matching on the server name if the master only dumps the whole
vldb if it needs to do pattern matching? That way some control exists
on the size of the vldb request. A different idea: is it possible to
pattern match on the master's, possibly out of date, idea of which
servers exist (and warn of the possible out-of-dateness)?
Are we having a meeting today?
To answer Jonathon's questions, the backup dump command (for any
dumpset!) uses the VL_ListEntry call, which is an iterative call for
returning one volume at a time from the vldb. This produces a large
number (one for each volume) of very simple vldb calls as opposed to
one large call to dump the whole vdlb. Which is preferred load-wise
is a matter of debate.
--Ted
------- Forwarded Message
Sender: jweiss@MIT.EDU
> As to the server, pattern matching on the server name (which matching
> all servers would have to do), requires dumping the whole vldb in one
> request (limitation in the AFS vldb code) to be completely robust.
> This is possible, but I would not want to have the master do such an
> operation each time it wants to expand a dumpset specification. So it
> was decided in the interests of load to require that at least the
> server name be explicitly given. This struck me as reasonable.
I'm concerned about our ability to do an archival dump of all
X... volumes. How do we do this now? Do we dump the entire vldb? We
should make sure that there is some (relatively easy) way to do this.
Jonathon
------- End of Forwarded Message
