Coda File System

Re: replicated backup :-)

From: Ivan Popov <>
Date: Sat, 15 Mar 2003 22:58:34 +0100 (MET)
Hello Jan,

> On Sat, Mar 15, 2003 at 05:30:56PM +0100, Ivan Popov wrote:
> > and that we still lack multilevel backup volumes (despite that they would
> > fit very naturally, but also afs and dfs people missed it).
> Ehh, we have had multi-level backups working reliably for almost three
> years now.

I think you have misunderstood me.

I mean that we should be able to have many "generations" of backup
volumes, say
"yesterday's files", "last week's files", "last month's files"

then a user would be able to access old versions of her files without any
"restore" procedure, just using appropriate paths
as a trivial example only:


~/../home.0.backup0/file     (on server #0 it is created every night)
~/../home.0.backup1/file     (on server #0 created each Sunday)
~/../home.0.backup2/file      ...

~/Yesterday -> ../home.0.backup0
~/LastWeek  -> ../home.0.backup1
~/LastMonth -> ../home.0.backup2
~/LastYear  -> ../home.0.backup3

Then ~/LastWeek/file gives me the state of the file as it was last Sunday

The backups of the same volume would share the same data and not waste
space in case they contain unchanged files.

By creating the backup volumes on multiple servers we would create the
redundancy almost the same way as with the online read-write data.

The whole concept is that all data is virtually always available online
without extra procedures. In Sun's SAM-FS the data is also moved to
inexpensive slow storage with time, but Coda does not need it, as even if
all server storage is slow, caching makes most of the file accesses fast.

Received on 2003-03-15 17:00:37