[CentOS] Re: Planning Mail Server (with low resources)

Bryan J. Smith thebs413 at earthlink.net
Wed Dec 7 23:38:21 UTC 2005


Rodrigo Barbosa <rodrigob at suespammers.org> wrote:
> I'm already sold for LTO-3 after this thread.
> I just need to find some good supliers.

It depends on your network.

For a server or two with less than 200GB to go off-site maybe
a few times a month, and no more than 12 tapes/year, then
LTO-1 is just fine.  I didn't realize the drives had dropped
to under $1K and the tapes to just over $25.

If you have a larger network and want to move several TBs
off-site per month -- especially if you're already spending
$2-3K just for the on-line backup server with TBs of disk,
then definitely add an LTO-3.  You'll thank me.  ;->

In either case, it's clear that if you're going to do tape,
LTO is now the only viable solution.  Unless you have legacy
AIT, DLT or something else, go LTO -- faster, cheaper,
better.

> Never used a 1" hard drive, so I can't really comment on
> that one.

I have a 5GB Seagate CFlash.  It's much faster and last a lot
longer (write-wise) than a traditional CFlash EEPROM.  I
always put in a FDD+9-in-1 reader in any system I assemble.

> Actually, I was talking about brand/vendors.

Like?

> Disks are good for (partial or full) HA, not backup.

I totally disagree.

Fixed disk is _ideal_ for near-line backup.  That means the
disks stay in one place and they are either always managed,
or at least powered regularly.  You should _always_ backup
from the end-nodes to disk over a network, _never_ directly
from end-node to remote tape.  You'd do this by sync'ing
differences of the end-nodes against the local store of the
last backup.  I.e., it's an "always diff" backup method.  The
best near-line disk backup maintains several recent volumes.

Tape is _ideal_ for off-line backup.  Ideally this means
being fed from an near-line disk backup that is local to the
tape drive.  That way you can drive it 24x7x365, and generate
*0* network traffic.  And you typically only do an "always
full" backup when it comes to off-line tape.  In an
enterprise environment with a half-dozen plus servers or
more, there is little reason to have more than one tape
drive.  You're money would be better spent on a dedicated
on-line/near-line disk backup solution with one tape drive.

> Some people call RAID a  backup solution, and that is where
> mistakes happen.

Or even snapshots for that matter.

RAID and snapshots are on-line.
Fixed disk is near-line. 
Tape is off-line.

In reality, you need _all_ 3.

People dismiss tape because they don't realize that it's just
_not_ good for near-line.  People dismiss disk because they
don't realize it's just _not_ good for off-line.  The
combination is not only necessary, but _complementary_.

> For backups, use tapes.

For _off-line_ backups, use tape cartridge.
For _near-line_ backups, use fixed disk.
For _on-line_ backup, at least use RAID, if not snapshots.

> I see now your line of thinking, but I don't agree with
> your terminology.
> Yes, I think tapes and disks are complementary for an
> avaliability (contingency, if you will) solution.

Fixed disk is ideal for near-line recovery.

Fixed disk allows you to do "always differential" network
backup.

Fixed disk is also ideal for feeding as well as verifying or
restoring tape cartridge.

Tape is ideal for off-lining data only when you need to.

Tape should be an "always full" backup, and _only_ done
locally to where the data is at -- _never_ over the network.

There is a great symbios between near-line and off-line that
people don't see.

They get caught up in the fixed disk v. tape -- either people
who  don't realize that they only hate tape because it sucks
at near-line and people who are too focused on disaster
recovery from off-line that they fail to see how much better
fixed disk is for near-line recovery.

> Your solution for a centralized backup server is one I
> would recomend too, and have even implemented it on the
past.

But how?

Are you streaming from end-node to end-tape over the network
during the limited backup window?

Are you at least buffering the process, although that still
has some backup window constraints?

Or are you only sending diffs to entire stores on fixed disk,
and then only sending full backups to tape only when you need
to off-line it?



-- 
Bryan J. Smith                | Sent from Yahoo Mail
mailto:b.j.smith at ieee.org     |  (please excuse any
http://thebs413.blogspot.com/ |   missing headers)



More information about the CentOS mailing list