[tlug] vmware, linux, snapshots

["Stephen J. Turnbull" <stephen@example.com>]

Darren Cook writes:

 > Are there any good vmware options here? Or is running rsync regularly,
 > to copy from main to backup server, the best choice?

Depending on exactly what the access pattern is like, Coda and Unison
might be possibilities.  Unison keeps two filesystems in sync, but
that's about all I know about it, I don't even know the protocol
involved.

Coda is a network file system, and works best in a true client-server
configuration.  For performance, information is cached on the client
and sent to the server.  With multiple servers, the information is
sent to and stored on all of the servers in a cell.

The catch is that for a file to be opened on the client, it must be
copied *entirely* to the client's cache from the server.  This means
that if you're dealing with DVD images or the like, the delay between
*requesting* a file and anything at all happening is huge.  And
obviously hosting a RDBMS on Coda is right out.  (OTOH, git repos and
Coda get along *really* well.)  There are technical reasons why this
is very difficult to avoid, so if the Linux servers are likely to be
frequently requesting new large content from the Coda server(s), or in
the event of request for new content, the reply must be realtime, Coda
is not for you.  Also, typical cache sizes are measured in hundreds of
megabytes; all frequently accessed files should fit in the cache.

Another drawback is that Coda does require some effort of the admins,
relative to simple rsync scripts or the like.  Configuration and
tuning of a Coda cell requires a fair amount of attention at first.

A third disadvantage is that the Coda client requires a special
kernel.  For active open source like Linux and the *BSDs this is not a
problem, but you're completely stuck waiting for the guys with the
Windows source license for Windows upgrades, and even the Mac doesn't
get a heck of a lot of attention.

OTOH, it's wonderful for mostly static, write-the-file-once content,
including things like Maildir MUAs or (some) IMAP servers.

There was some work done on "huge cache" configurations (10s of
gigabytes), so that effectively all work is done on the Coda client,
and you have online, "hot" backup to the Coda server.  But I don't
know if that ever was robust.

Note that if you have large new content added to the client, there
will be a *lot* of traffic to the server for a while.  So you're
likely to get thrashing if you have both an active Coda client and a
Coda server running on the same host.  You really do need to have the
Coda client (which might host a webserver or something like that, as
well as user home directories) and the Coda server (which should do
nothing but Coda) on separate hosts in a production configuration.

HTH