Anthony Worrall <[EMAIL PROTECTED]> writes:
> >From: "John R. Jackson" <[EMAIL PROTECTED]>
> >
> >>Using a NFS mounted holding disk doesn't seem possible ...
> >
> >I would consider that a feature :-). Why in the world would you drag
> >a bunch of dump images across the network to an Amanda server and then
> >send them back across the network, using NFS of all things, then turn
> >around and drag them back a third time to finally go to tape? Ick.
> >
> One reason would be the NFS server has a 1Gb interface and the clients and
> tape server have only 100Mb.
Hi,
I'd like to add a positive touch to this thread and suggest other setups:
1) NFS-mounting of everything to the NFS-Server in a "one client setup"
So data is copied once from non-amanda clients to the nfs-server. The
spindle parameter in the disklist can be used to control network
congestion (e. g. dumping one non-amanda client in a collision domain
at a time). I guess that the 1Gb/100MBit-Hub is fully
switched. The drawback is that the nfs server needs to do all the
software compression (if used at all). The holding disk in this setup
is still on the tape (and amanda) server.
2) Remote tape
Make the nfs server the amanda server and use a remote tape device
(e.g. /dev/rmt0 for Linux). I'd like to hear comments if amanda
supports that. Clients can use their own horsepower to do software
compression. Performance for this setup depends on keeping the tape
drive streaming. The speed advantage of a local holding disk might be
gone, but a slow tape, clients not eating up all bandwith and a fully
switched network might make this possible.
3) Move the tape to a central position
and use the nfs server as amanda server with a local tape. I'd use a
dedicated holding disk to maintain NFS disk performance during
backups.
A related idea: Hierarchies of holding disks
This can make a lot of sense to backup remote company outlets with a
fast LAN and a slower WAN to a centralized amanda server. One holding
disk per outlet allows for consistent data backups during the night to
the outlet's holding disk. You can then make use of traffic shaping on
the WAN router to gracefully move the data to the headquarters amanda
server during the next 24 hours.
At the moment I'd implement that via a two stage setup: Each outlet
has its own amanda server (with local clients) and no tape drive. The
holding disk of the local configuration is backed up to the main
amanda server (with tape drive). Restores are a painfull two stage
process and involve sending the whole holding disk of that day over
the WAN.
Implementing this directly into Amanda would need more than the
rewrite for Amanda 2.5, as we use "relay servers". This resembles to a
certain degree the way we do SMB backups. Do some commercial products
(like Tivoli or Legato) support hierarchies of holding disks?
