Sorry for being a bit late in replying to this.
As you indicate, distributed processing ought to be more efficient, depending on the capabilities
and loads of respective servers. I typically use client side compression, because all the servers
I'm backing up have sufficient resources to do it themselves, and that would reduce the load on the
server running Amanda. However, it is also the case that far and away the bulk of my data is on the
servers that are running Amanda. They have the external arrays with tens of TBs of data. So, local
compression in that case is actually on the Amanda server.
Having the bulk of the data and the Amanda backups on the same server has some advantages. It means
that the path from the Raid array to the holding SSD to the LTO6 tape is all direct over SAS, and
the SAS connections are all separate. That is, the SAS card that the arrays are attached to is
different from the SAS card that the LTO6 is attached to, so they don't compete for bandwidth. The
SSDs are internal, so they are on a separate connection. Then the trick is to have adequate CPU with
enough cores to have a mess of processes running for Amanda without causing a slowdown of the other
services that are running on that server.
I could imagine a case in which a dedicated Amanda server with sufficient cores might be doing all
the compression for a number of other servers that were too busy with their own services to manage
the compression of the backups. I just don't have that case.
The only issues I've run into with recovery were with a weird version of gnutar that had issues with
Amanda. The solution was to make sure I was using a compatible gnutar and to test recoveries early
in the installation/setup process so that I know the combination I have actually works.
On 11/1/16 6:17 PM, Ochressandro Rettinger wrote:
So, I understand the subject of software vs: hardware compression now (thanks for
that very enlightening description of why it’s good to let Amanda handle the compression, Chris)
but now I’m wondering about the question of server vs: client compression. I would tend to think
that it would be best to use client compression, if possible, because then you can take advantage
of the parallel nature of all of those separate processors working on their compression tasks.
But from what several people have said, it seems like server side compression is a common choice.
Why would I want to choose one over the other? (Beyond the fact that in my
particular case, client side compression isn’t actually working for recovery purposes.)
Thanks for answering all these questions, guys. :)
-Sandro
--
---------------
Chris Hoogendyk
-
O__ ---- Systems Administrator
c/ /'_ --- Biology & Geosciences Departments
(*) \(*) -- 315 Morrill Science Center
~~~~~~~~~~ - University of Massachusetts, Amherst
<[email protected]>
---------------
Erdös 4
