Amanda-Users

Re: dealing w/ huge data backups

2005-06-03 06:23:44
Subject: Re: dealing w/ huge data backups
From: Alexander Jolk <alexj AT buf DOT com>
To: Cameron Matheson <cmatheson3 AT yahoo DOT com>
Date: Fri, 03 Jun 2005 12:08:40 +0200
Cameron Matheson wrote:
many systems should be able to fit on one tape
(i'm using hw compression on tapes that hold a little
over 100GB w/ no compression).  Anyway, the problem
now is that i need to back up a system  that is
120GB.... i can split that up no problem (i'm just
using tar anyway), but the question is would it be
better to just bump up the 'runtapes' parameter to 2
or 3 tapes,

I'd like to give you a little stronger advice than the others: you *definitely* need to split this one up. Really.

Just bumping runtapes will help you if you total backup size is larger than one tape; but it won't help you at all if a single DLE is too large to fit on one tape. amanda never splits a single DLE across tapes (barring patches to the 2.5 series which I don't think you are using). So essentially, whether or not you bump runtapes is a different decision, but it won't help you for this huge DLE.

Now what happens is that amanda will most certainly finish dumping this DLE rather late during her run, when the tape has already had a certain number of dumps written to it. You could try to fiddle around with dumporder and the number of dumpers, but I'm sure that's really not worth it. Anyway, there's a fat chance that amanda would try to tape the dump, hit EOF, and start over on the next tape. For a huge DLE, that can mean tens of GB of wasted tape, along with the corresponding time.

Now just split up your DLE in lots of smaller units. By lots I mean easily 10 or 20; I normally try to keep my individual DLEs below 10GB. With this setup, when amanda hits EOF and retries on the next tape, you'll never have more than 10GB of wasted space, and a lot less quite often. And wait, there's a second advantage: amanda will schedule level 0's for different DLEs at different times. If you have a huge DLE, its level 0 will take several hours to run, once every dumpcycle. If you have lots of smaller DLEs, a few of them will get a level 0 every day, others the next day, and so on.

Now for the actual splitting-up, if you need help or you want to run your config by the list, feel free. I use a home-grown perl script along with `du -xks *' output to split my DLEs, others do it by hand, it all depends. Just for example, here's what a properly split huge disk will look like:

# draco:/raid2
draco   /raid2/PROJECT  svcomp-work-tar 1
draco   /raid2/DESC_SCANS       svcomp-final-tar 1
draco   /raid2/GENERIC  svcomp-work-tar 1
draco   /raid2/GAME     svcomp-work-tar 1
draco   /raid2/SPRITE   svcomp-work-tar 1
draco   /raid2/TMP      svcomp-work-tar 1
draco   /raid2 {
        svcomp-work-tar
        exclude append "./PROJECT"
        exclude append "./DESC_SCANS"
        exclude append "./GENERIC"
        exclude append "./GAME"
        exclude append "./SPRITE"
        exclude append "./TMP"
        } 1
# end draco:/raid2

Alex


--
Alexander Jolk         /         BUF Compagnie
tel +33-1 42 68 18 28 /  fax +33-1 42 68 18 29


<Prev in Thread] Current Thread [Next in Thread>