* John R. Jackson <[EMAIL PROTECTED]> (Mon, Jul 30, 2001 at 05:04:47PM -0500)
> >Does anyone know if that puppy is supported by amanda ?
> If it's SCSI, Amanda can do it (or it should be able to so we'll make it).

It is,
the thing is that it's a taperobot/tapechanger, and I remember reading
discussions that getting that to work is not always a trivial issue.

And Ill likely only have a week to set it all up (and do a whole lot more)
so I'd prefer not to have to spent too much time tinkering with the scsi
changer configs ;)

>>Does anyone have any recommendations for backing up this amound of data ?
> Just a couple of general theories:

>   * Make sure no one "disk" (disklist entry) gets out of line.
>     Ignoring the tape overflow issue, it would still be a bad thing
>     to have 1.9 TBytes of your 2.0 TBytes in one file system.  The day
>     Amanda decides to do that full dump during the dumpcycle is going
>     to be bad every time it rolls around (we have one config like that
>     here and it's a nuisance).

Hence gnutar and calcsize ;)
seriously, it's gonna be a bunch of linux boxes each with 100 - 150G of
diskspace (2 IDE drives of 50 - 75G)

>   * Extend your dumpcycle so Amanda can spread the full dumps out.
>     Yes, that means more tapes.

>   * Run in parallel (yes, more drives and more tapes).  We backup up
>     well over 2 TBytes and this (plus a 28 day dumpcycle) is how we
>     survive.

What kind of tapedrives do you use ?

>     If you don't have enough clients (i.e. they are silly enough to put
>     the whole 2 TBytes on a single server), I have a minor patch to let
>     more than one amandad run at a time so you can have multiple tape
>     servers working at the same time.

Do you need a patch for that ?
Im running 3 amandads on the same machine every night
(all 3 to different tapedrives ), and each client is running multiple
clients (or at least can run ) at the same time.

>   * Do the math :-).  You're familiar with Amanda and know how it behaves.
>     Write up some scenarios for splitting up the data and do rough
>     estimates of the timing.  Then play with the variables (dumpcycle,
>     holding disk space, number of configs, drive speed, etc).

Yes, 
I'll have to do that anyway 
Ill need to make a cost/benefit analysis of the various scenarios .


        Gerhard,  <@jasongeo.com>   == The Acoustic Motorbiker ==       
-- 
   __O  If your watch is wound, wound to run, it will
 =`\<,  If your time is due, due to come, it will
(=)/(=) Living this life, is like trying to learn latin
                in a chines firedrill

Reply via email to