[Pc_Support] Re: Backup methodology -- centralize for lower TCO

[Bryan J. Smith]

On Tue, 2006-06-27 at 13:54 -0400, Damien McKenna wrote:
> That's what I'm trying to avoid, hence my questions on how to do it :- 
> P  My question was a specific implementation question, i.e. should I  
> first backup the files to disk and copy that archive file to tape, or  
> should I do a separate backup of the data straight to tape?

But to _what_ disc?

> The problem with the former, as I see it, is that you end up with an  
> archive within an archive, i.e you're not backing up the source files  
> you're backing up a 150gb file and that's what you'd be restoring  
> should the need arise.

Actually, the software shouldn't do that.  It will "unarchive" the
source file into a tape stream.  No double archive.  However, if the
software is _not_ intelligent, then you're just putting a full plus
separate incrementals to tape.

But to _what_ disc?

> We have one centralized backup server running BackupExec, I don't see  
> the purpose of having NetBackup.

Far more flexibility.

> Needless to say the Symantec site  
> is pretty useless in explaining in real terms how each one works  
> versus the other, and the rep at CDW said we didn't need NetBackup to  
> do basic stuff that we just need the "Advanced File Backup Option"  
> for BackupExec.  It seems to be the usual "enterprisey" BS to me -  
> there are technical issues at hand but its so wrapped up in BS PR  
> terms and marketed to management that the lowly geeks can't get the  
> information they want/need to make the decision.

Oh, if you're using "Advanced File Backup Option," then it may do some
of what you want.  NetBackup has more options, although most people
don't take advantage of them.

Heck, most people still stream directly to tape today, over the network
(yikes!).

> You misunderstood.  We already have the LTO-2 drive I've mentioned  
> before, I don't want to go SCSI for the near-line portion of our  
> backup routine.

Then just throw some local 3Ware storage on your server and you're
good-to-go.

BTW, what is your server?  I/O?  What is its disk as well SCSI
controller for the tape?

> Right now I'm looking at a SCSI enclosure for $55 (http:// 
> www.newegg.com/Product/Product.asp?Item=N82E16817145006) and a $20  
> card (http://www.newegg.com/Product/Product.asp?Item=N82E16816132002).

Huh?  Why not just get an eSATA cable converter and use the on-board
SATA channels?  Or is the server that old?

If so, maybe that should make you consider spending money on a newer
backup server that can handle the stream load?

Furthermore, be _careful_ in plugging in 32-bit PCI cards on servers
with 64-bit PCI-X slots.  You can't use 5V cards and you will often
_slow_down_ the PCI-X slots to 33MHz.  Just FYI.  ;->

> Implementation time isn't a concern, its daily maintenance that's the  
> problem.

And that's where removable drives suck, _hard_.  ;->

The only thing you want to make "manual" in your backup strategy is when
you take the off-line media (e.g., tape) off-site.

3.5" disc is not like tape cartridge.  At a _minimum_, I recommend a
2.5" disc.  They take 10x the Gs.

There's nothing worse than a server that has an external disc going
"hiccup" ... "hiccup" ... "hiccup" and it's _not_ RAID so the _whole_
system starts _crashing_.

Think about it.  ;->

At Boeing, we had 2 labs with Linux and Windows having constant
_crashes_.  The problem?  External USB 2.0 and FireWire drives.  They
even tried removable SCSI -- same issues.

I put in a $100 3Ware Escalade 8006-2LP card with two 3.5" bays.  And in
that 3.5" bay, I put 2.5" drives that were rubber-standoff padded.  At
no time was more than 1 disc removed.  That prevented the "downing" of
the server, as well as the reliance on the disk not to have issues.

But still, there was _manual_overhead_ in removing drives, telling the
3Ware software to rebuild, etc... when drives were removed/added.  It
_not_ ideal and the _only_ reason I did it that way was because we
couldn't centralize the backup for those labs because they were
independent (i.e., classified).

In your case, your servers are all on the same network.  So in your
case, you should have 1 backup server with 1 tape drive -- period.

> Two main servers (the file server itself and a second server running  
> MS Exchange) plus a negligible quantity of remote files around the  
> network.

In that case, you're already an ideal candidate for a centralized backup
server with local, near-line store.

> I don't see the room for error.

Sorry, the sarcasm didn't come through.  ;->

What I meant was that anytime you're doing "manual" tasks, you've got
lots of places were errors can (and will) occur.

> They won't be flying around?!

Again, you didn't get what I meant at all.

If you're pulling drives and moving them around, plan on 50% of 3.5"
drives _failing_ per year.

> At what point did I say that we *weren't* doing a centralized backup?

You were talking about getting removable/portable hard drives.
I was advocating _against_ that.

> Won't happen.  We have a DL380 file server running 140gb of space on  
> a SCSI RAID-5 with an LTO-2 tape drive attached and currently using  
> about 60-70gb of space.  We have a second DL380 with <20gb of  
> Exchange 2003 data.

I'm missing something here.  You have 2 file servers and countless, file
backup from individual systems.

And why can't you spending $1,000 on a dedicated, 500GB backup server?
Specially when they'll spend _more_ on man-hours moving around and
manually backing up with removable drives?

> I know and don't care, it is out of scope for this project.

So _what_ are you going to do?
I'm scratching my head on that.  

You're going to go around with removable/portable drives?
And how is _that_ going to save you money?

Especially if you have to "down" the server to remove disks?
Or spend $100-300 on a 3Ware or SAS card so you can hot-swap SATA?

Why not spend a little more money and put in a _real_ backup server for
about $1,000 with 500GB of disk?  Seriously -- put in a $500 system with
a $100 3Ware Escalade 7006/8006-2LP and two $200 500GB ATA/SATA drives.

> More specifically I want:
> 1. Full backup of the servers in question to a local disk (via  
> external SATA), updated weekly.

And how long do you expect that SATA drive to last, "flying around"?
Or how many SATA drives are you going to "buy" to do such?
And how often have you budgeted their "replacement"?

$1,000 adds up really fast.  ;->

> 2. Full backup of the servers in question to LTO-2 tape, rotated weekly.

>From one to the other, over the network?
Over a dedicated channel?

> 3. Daily deltas of the servers in question to the same local disk.

So if that server blows up, or its filesystem tanks, you lose as well?

> In effect I want #1 and #2 to the same.

I'm not getting what you mean.

> BackupExec seems a bit dumb, er, crap, in this regard.  Yes, you can  
> create delta backups but it doesn't seem to let you out-of-the-box  
> create a rotating group of disk backup files.

I know.

> So, to make sure you're not getting mixed up in the month-long  
> conversation...
> * We've got about 160gb of data spread over two servers.
> * Both servers are the same - DL380's with SCSI drives in RAID-5.
> * Both servers run Windows 2003 (we upgraded recently from '2000).
> * One server, with the bare files, has an LTO-2 drive connected to  
> the internal SCSI controller.

What speed?  LTO-2 wants Ultra160.

> * The other server just has MS Exchange 2003 with some related  
> utilities.
> * There's a tiny amount of workstation-level data that's being backed  
> up but I can use something on that specific PC if needed, i.e. it  
> isn't really part of the equation.

But it _does_ take up your time?
And that requires effort to address.

> * I want to do a weekly full snapshot of both servers.
> * I want daily deltas of both servers.
> * I want to copy the weekly snapshot to tape so I have a 1-to-1  
> relationship between the on-disk weekly snapshot and the tapes.

Why not make your local disk a _superset_ of backups to tape?
So you can go "back in time" 1, 2, 3 ... 6 days, or even 1, 2, 3
weeks ... maybe even another 2 months?

> * The weekly tape backups should be of the original data, i.e. if I  
> do a list of the tape's catalog I want to see the source files and  
> not a 160gb backup file which must be extracted in its entirety  
> before I can then access the source files.  This seems to be a  
> problem with many solutions.

Huh?  I'm not following you.

> * At the hardware level we're looking at an external solution due to  
> space limitations.

But how are you going to "hot-swap" those external solutions?

> * SCSI is too expensive for us, as is SAS.

I know.  That's why I said a 3Ware solution would be _cheaper_, while
still giving you "hot-swap."

> * External SATA of some sort, be it official e-SATA or just an  
> external SATA connector, seems to be a good route to take in terms of  
> speed vs reliability vs cost.

But how are you going to "hot-swap" it?

> * I'm looking at either a Western Digital RE-series or Seagate NL35  
> drive for the improved reliability over consumer grade drives.  $85  
> for 250gb is plenty I think, or 320gb for $120.
> * I'm looking at about $20 for a controller card and $55 for a case,  
> so $160 to $195 total.
> * BackupExec by default has crappy D2D2T support, it seems a $380 add- 
> on is required to do what I want.
> * Our licenses for BackupExec are up for renewal so it is worth  
> considering alternative products.

In the famous words of Vern, "I must be an idiot" because I'm trying to
figure out how you're going to spend less than $1,000 and why you keep
saying my suggestions are "outside the scope of the project."

Especially since my suggestions _completely_eliminate_ the manual
processes you talk about -- other than when you write to tape.  ;->


-- 
Bryan J. Smith           Professional, technical annoyance
mailto:b.j.smith at ieee.org     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.