Poll; Tape drives

[Lennart Sorensen]

On Fri, Mar 07, 2008 at 11:27:38AM -0500, Madison Kelly wrote:
> Off-list I was speaking to a friend who works in IT at a big company, 
> and she also voted for tapes, saying that in large scale environments 
> "it's the way to go".
> 
> It surprises me, in a way. I was expecting to hear grumblings that tape 
> was a necessary evil, but apparently it is still a preferred solution in 
> larger situations.
> 
> As Len mentioned though, up to a (fairly large capacity) point, Disk 
> still makes more sense. So then, I need to ask;
> 
> Q. What raw storage capacity required is the "break point" where tape 
> becomes preferable to disk storage?

Well tapes do have the one advantage that you can easily remove them and
bring them offsite for storage, which is a little harder with disks
although usb enclosures work great for small backups in that manner.

I would say for off site storage the choices are:
external disk for backups less than 2 or 3 TB (at this time).  I suppose
having to unplug and move a harddisk versus a tape may not be that big a
deal so maybe if doesn't make much difference.

The real answer to off site backups is to do them over a network link
(internet or dedicated) to an archive system at another location, which
could be disk or tape based.

Tapes are a bit more robust when carried though since they don't have
the same fragile electronics and disk heads and such inside.  I would
still recommend the remote backup method instead, in which case
cost/capacity is the real issue.

> If I can answer that, then I can start to decide if people who need that 
> capacity are realistically in the range of potential users of my modest 
> program.

Well the cheapest I found for tapes so far seems to be about $100 for
800GB, which is then 12.5 cents per GB.  500GB disks can be had for
$100, or about 20 cents per GB.  So if we call that 8 cents per GB
advantage for the tapes, but you have to pay for an expensive tape
drive, and tape changer with lots of slots, and deal with a much larger
latency of access (which for an archive should not be important for the
user anyhow), as well as potentially expensive software to manage the
tape changer (compared to just storing stuff in a filesystem managed
however you want).

Apparently you can get a 700 slot tape library for about $85000, and
then you can add tape drives to it for about $5000 each (SDLT or LTO or
whichever).  Assuming access speed is not important and we only put in
one drive then we have $90000 for the tape changer.  Adding 700 tapes at
$100 each (I imagine you can get a bulk discount somewhere) for $70000
in tapes, and we get $160000 for a total size of 700*800GB which is
560000GB capacity, at 28 cents per GB.

Building a computer to hold the required number of 500 or 750GB drives
(call it 1000 drives) would be quite a task, although multiple smaller
machines would still do the job together.  I have certainly seen a 5U 24
hotswap SATA bay cases where a 24 channel raid controller would work well.
It would of course include a ton of processing power which may or may
not be needed, although you would need some system to manage the tape
changer too.  I found one place that will sell a 5U box with 24 1TB
drives in it for $13000 which is 54 cents per GB, but includes lots of
ram and cpu power too.

Using the google model of lots of generic small machines you could do a
simple desktop box for $400 with 8 750GB sata drives giving you 6000GB
storage for $1600 or 26 cents per GB.  It would mean managing a network
of close to 100 small boxes, although you might find a use for that
amount of processing power too.  It does come out to about the same cost
per GB as the tape library though, although it would use quite a bit
more electricity to operate.  I suspect the point where it makes sense
to go to tape is probably when you hit a tape library with 100TB
capacity or more as long as the rather slow access to archived
information is just fine.  Many archives that size do have disk based
caches to help speed up access to information when it is requested
multiple times in a short timespan.

I think you need a library large enough that you need management
software designed to manage an archive, which I suspect is not what you
are working on.

--
Len Sorensen
--
The Toronto Linux Users Group.      Meetings: http://gtalug.org/
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists