Thanks for that.
I don't think it's our intention to backup ALL 10TB nightly. And I
would expect we spread out the level 0s like you mentioned.
Any other thoughts, that people have on this would be appreciated
Thanks.
Matt.
Chris Hoogendyk wrote:
On 8/22/11 11:17 AM, Matt Ingram wrote:
Hi All.
We're looking at using amanda for a virtual tape library.
We have two sites, and four SANs. here's a basic outline:
Site A
=====
ProductionA SAN (5TB)
"Backup of ProductionB" SAN (27TB)
Site B
=====
ProductionB SAN (10TB)
"Backup of ProductionA" SAN (17TB)
As per the above, we want to backup our data offsite. Looking for
some thoughts on how we should implement this. Some best practices,
some gotchas, etc would be great.
Some initial questions are, where should we place our amanda backup
servers. We're thinking we need one for each site, but is it better
to have the amanda server local to the data it's backing and have the
LUN for the backups remote OR, have amanda server remote to the data
and have the LUN for the backups local.
Would greatly appreciate any thoughts and info all you amanda experts
have :)
Hmm. 10TB over WAN? You can calculate what that would take, depending
on what your connections are. Suppose you had a 1Gbps connection
between the remote sites. That's 125MBps, which is 7.5GBpm, or
450GBph, or 10.8TBpd. But, you never get 100% utilization of rated
speed. So, figure if you had a 1Gbps connection it would take a couple
of days (with no competition for the bandwidth) to get a full backup.
Now, plug in what you really have for bandwidth, what portion of that
bandwidth is in use for other purposes, and project how long it might
take.
From that, figure out what portion of the whole would make sense to
try to do a full backup of at one shot. Then divide up your total
storage into enough pieces (DLE's) so that none of them are larger
than that. Then come up with a plan that sets your dumpcycle to a
number of days so that you'll be able to spread the fulls out over the
dumpcycle and actually accomplish them within a single run along with
incrementals of all the other DLE's.
That's sort of a starting point for thinking about it. Once you have
details about how much actual data is on those systems, what your
bandwidth is, what the current utilization of the bandwidth is, and
how fast/frequently your data changes; then, I'm sure there will be
more concrete/detailed suggestions for how to handle it.
--
Matt Ingram
Intermediate Unix Administrator, IS
Canadian Bank Note Company, Limited
\m/
