Re: [SLE] Backups -- near-line disc and off-line tape
- From: "Bryan J. Smith" <b.j.smith@xxxxxxxx>
- Date: Wed, 28 Jun 2006 22:35:03 -0400
On Wed, 2006-06-28 at 18:51 -0700, suse@xxxxxxxxxxxxxxx wrote:
Looking for suggestions for backups:
Current situation:
3 data servers: daily tars copied to a 4th server with a tape
backup running at night.
That's actually a good strategy -- you buffer to disc, and then disc to
tape. I *NEVER* stream directly from a system over the network to tape.
In fact, such real-time tape backup is why most people hate tape.
The ideal setup is to use disc as your near-line store (for quick backup
with immediate restore), and tape as your off-line store (for taking
backups off-site). In fact, using that strategy, you only really need
to do a tape backup when you actually take backups off-site -- e.g.,
weekly.
The most ideal setup is when you have enough near-line storage to
accommodate a _full_ backup of _all_ your servers. That way you only
need to send incremental deltas every night from servers to that backup
with near-line storage. From there, at your convenience, you can decide
what to put from the near-line disc to off-line tape (or not).
You don't need SCSI drives for this backup system, since your discs will
be idle most of the time. A 3Ware 4-channel SATA card with four (4)
320GB WD3200SD** (24x7 rated near-line enterprise commodity disc**)
would give you 1 TB usable RAID-5 for $600. You can then put that in a
new $500-600 backup server, or just use an existing server (that doesn't
have much load at night for the deltas to near-line disc, and weekends
for the write to near-line tape).
For a more complete discussion of the evolution of legacy real-time tape
backup to buffered disc-to-tape backup to today's combination of
near-line disc with off-line tape and the final evolution of the
"virtual tape library," see 2005 September Sys Admin:
"Dissecting Virtual Tape Libraries"
http://www.samag.com/documents/sam0509a/
The article is purposely written from a _sysadmin_ functional concept
and viewpoint, and _not_ a network/storage admin viewpoint.
[ **NOTE: Pay the $20-30 "premium" for the Hitachi "T7K", Seagate
"NL35" or Western Digital Caviar "RE" series. These are commodity discs
that are _rated_ for 24x7 "near-line" (sometimes, and quite
_in_correctly, called "RAID" rated). In a nutshell, they are the
Hitachi 7K, Seagate Barracuda ATA 7200.x and Western Digital Caviar SE
products that test to _higher_tolerances_ than their consumer/retail
rated models. They are well worth it. I'm partial to the WD3200SD
(107GB/platter, 320GB Caviar RE, _not_ RE2) because it is very proven as
a reliable model (even in its consumer WD3200JD version too). _Never_
use "brand name" to base a decision -- as even hard drive vendors
outsource these days. E.g., Seagate Barracuda ATA 7200.8/9/10 models
are outsourced to Maxtor, and not as reliable as their prior 7200.7
series, so it didn't surprise me when Maxtor and Seagate merged just
recently. Some of the newer Western Digital models (like the 500GB)
aren't the same, sound design -- and Hitachi has gone back to using
5-platters in their 500GB disc (the last disc to do that was the IBM
Deskstar 75GXP -- the Hitachi 500GB is also the Deskstar series ;-).
But at least with the Hitachi "T7K" (not the consumer 7K), Seagate NL35
(not the consumer Barracuda ATA 7200.x) and the Western Digital Caviar
RE (not the consumer Caviar SE), they _are_ tested to higher tolerances
and _rated_ for 24x7 "near-line" (i.e., not spinning 24x7, but still
okay to be powered 24x7) instead of the standard 8x5 "consumer"
tolerances. ]
The local PCs are NOT backed up, nothing is supposed to be stored there.
With an intermediate backup server, you can always choose to rsync
select directories to it.
BTW, a popular strategy for backup servers is to maintain "snapshots" of
backups. Using kernel 2.4 with LVM**, you can take daily and weekly
"snapshots" of the backup server, and then serve those "snapshots in
time" back out (read-only) via NFS and SMB.
E.g., a typical strategy I personally use are the last 6 days and then
the last 2 Fridays before those last 6 days (3 weeks back) -- 8
snapshots. Every night I do the "snapshot" of the filesystem _before_
the servers and workstations do a rsync to send their deltas of what has
changed in the last 24 hours. From that, users can go back and look at
_any_ backup via NFS or SMB from the last night, any of the previous 6
days, or the 2 Fridays before last.
**NOTE: kernel 2.6 with LVM2 and DeviceMapper is still developing
snapshots (with write support and other advanced features). Red Hat has
been shipping the feature in more recent kernels, but most (including
myself) still don't trust it yet. So I'd stick with a kernel 2.4
distro.
The tape drive has died: a replacement is around $3,000. The
tapes were convenient because somebody could take them home next
night and rotate.
What technology? AIT? DLT? LTO? Other? You'll want to stick with
what technology you already had so you can read tapes from before.
If you don't need to read your old backup cartridges, then I suggest
going with LTO -- a multi-vendor standard that is the best
price/performance and also DLT-quality longevity (only IBM's proprietary
formats are longer lasting).
The backups are right now 40Gb, but could easily be reduced to
10Gb.
Don't do that to yourself. If you need 40GB, backup 40GB. Don't add
administrative burden to your job.
In reality, consider spending $1,200 on a new backup server with 1TB of
near-line (ATA-based) storage. If you don't need to read your old
tapes, LTO-1 (100GB/20MBps native, 240GB/48MBps 2.4:1 typical) starts at
$800 and tapes are under $25/each.
If you use a combination of near-line disc and off-line tape, you really
only need enough disc to store full backups (2x if you want to use
snapshots) and then enough tape cartridges to store 4 recent weeks
(rotated) plus one per month for long-term retention (if you need it).
That cuts down on your tape usage/need/costs.
What can you suggest so that somebody can take the backup home?
Since LTO-1 is so cheap, I recommend you stick with tape if you need
portability. Use disc for near-line (local, on network) and _only_ use
tape for when you need to take it off-line (like off-site).
A common (and _poor_) assumption is that disc can take "swapping
around." It can_not_. At the most, I'd by 2.5" discs and pad them like
made in a tray because they can take 10x the stock of 3.5" discs. But I
would _not_ rely on 3.5" portable discs. Especially today's 3.5"
commodity discs that are _not_ designed for staying off-line long.
Lastly, you have the "hot-swap" issue. SATA does not "hot-swap" well
unless you put it on a 3Ware or other _true_ hardware RAID controller in
at least hardware RAID-1 (long story). Now a 3Ware Escalade 8006-2LP
will only cost you $125, plus another few hundred for a pair of discs or
trays, but at that point, you're half-way to a LTO-1 drive anyway.
The key to having a _complete_ recovery solution is to:
A. Realize that you not only _can_ but _should_ use disc backup today.
Disc is a reliable mechanism, fast, random-access and can store "full
copies" you can "delta" against (instead of having to pass full copies
over the network every night).
B. Realize that you still need an off-line solution for off-site
portability. Since this solution only goes off-site every week or
month, it means it does _not_ and often should _not_ be your primary
mechanism of backup. You should be using near-line disc, and then only
commit from disc to off-line _when_ you actually need to take stuff
off-line/off-site.
C. Reality that disc is _not_ portable and has _poor_ tolerance to
vibration, at least compared to tape cartridge. Which is why you should
_not_ be using disc as your off-line/off-site, but tape. Now you could
opt to do Internet-based backup -- and you probably can do it with only
40GB. But it's also nice to have _multiple_ off-line copies, and not
just one. And that's where tape still has a great advantage.
Again, if you read the 2005 September article, it lays out all the
aspects of not only "getting stuff backed up," but looking at the
aspects of the _recovery_ from a feasibility/reliability standpoint.
In a nutshell ...
Tape sucks compared to disc when it comes to immediate backup and
recovery.
But disc sucks compared to tape when it comes to longevity, environment,
tolerance and portability.
The ideal setup is to leverage the two, as they complement each other
quite well.
--
Bryan J. Smith Professional, technical annoyance
mailto:b.j.smith@xxxxxxxx http://thebs413.blogspot.com
----------------------------------------------------------
The existence of Linux has far more to do with the breakup
of AT&T's monopoly than anything Microsoft has ever done.
--
Check the headers for your unsubscription address
For additional commands send e-mail to suse-linux-e-help@xxxxxxxx
Also check the archives at http://lists.suse.com
Please read the FAQs: suse-linux-e-faq@xxxxxxxx
- References:
- [SLE] Backups
- From: suse
- [SLE] Backups
- Prev by Date: Re: [SLE] How to use "man" command?
- Next by Date: Re: [SLE] How to use "man" command?
- Previous by thread: [SLE] Backups
- Next by thread: Re: [SLE] Backups
- Index(es):
Relevant Pages
|
