Joe Atzberger wrote:
On Thu, Aug 27, 2009 at 4:25 PM, Edward M. Corrado <ecorr...@ecorrado.us>wrote:
Nate Vack wrote:
On Thu, Aug 27, 2009 at 1:57 PM, Ryan Ordway<rord...@oregonstate.edu>
wrote:
$213,360 over 3 years
If you're ONLY looking at storage costs, SATA drives in enterprise RAID
systems range from about $1.00/GB to about $1.25/GB for online storage.
Yeah -- but if you're looking only at storage costs, you'll have an
inaccurate estimate of your costs. You've got power, cooling, sysadmin
time, and replacements for failed disks. If you want an
apples-to-apples comparison, you'll want an offsite mirror, as well.
I'm not saying S3 is always cost-effective -- but in our experience,
the costs of the disks themselves is dwarfed by the costs of the
related infrastructure.
I agree that the cost of storage is only one factor. I have to wonder
though, how much more staff time do you need for local storage than cloud
storage? I don't know the answer but I'm not sure it is much more than
setting up S3 storage, especially if you have a good partnership with your
storage vendor.
Support relationships, especially regarding storage are very costly. When I
worked at a midsize datacenter, we implemented a backup solution with
STORServer and tivoli. Both hardware and software were considerably
costly. Initial and ongoing support, while indispensable was basically as
much as the cost of the hardware every few years.
They can be depending on what you are doing and what choices on software
you make, but for long term preservation purposes they don't have to be
nearly as expensive as what Ryan calculated S3 to cost. If you shop
around you can get a quality 36GB array with 3 yr warranty for say
$30,000 that is almost $180,000 less than S3 (probably much less, I'm be
less than generous with my Sun discounts and only briefly looked at
there prices). Even if we use the double your cost for support, it is
still over $50,000 a year less for 3 years. Yes, we might need some
expertise, but running a 36TB preservation storage array is not a
$50,000 a year job and besides, what is wrong with growing local expertise?
...
Yes, maybe you save on staff time patching software on your storage array,
but that is not a significant amount of time - esp. since you are still
going to have some local storage, and there isn't much difference in staff
time in doing 2 TB vs. 20 TB.
There's a real difference. I can get 2 TB in a single HDD, for example this
one for $200 at NewEgg:
http://www.newegg.com/Product/Product.aspx?Item=N82E16822148413
<http://www.newegg.com/Product/Product.aspx?Item=N82E16822148413>
Any high school kid can install that. 20 TB requires some kind of
additional structure and additional expertise.
Well building a 20 TB storage device and getting it to work can actually
be very cheap and doesn't require a PhD (just a local GNU/Linux geek who
likes to play with hardware) if you are OK with a home grown solution. I
wouldn't be satisfied with that, but I don't see how a commercial
offering that adds up to $150,000 worth of expertise and infrastructure.
You may some time on the initial configuration, but you still need to
configure cloud storage. Is cloud storage that much easier/less time
consuming to configure than an iSCSI device? Replacement for disks would be
covered under your warranty or support contract (at least I would hope you
would have one).
Warranties expire and force you into ill-timed, hardly-afforded and
dangerous-to-your-data upgrades. Sorta like some ILS systems with which we
are all familiar.
Yes some application upgrades can cause issues, but how is that
different if your application and/or storage is in a cloud?
The cloud doesn't necessarily stay the same, but the part
you care about (data in, data out) does.
How do you know they won't change their cloud models? And you don't even
have a warranty with the cloud. They won't even guarantee they won't
delete your data.
As long as you use a common standards based method of storage, you won't
have any more issues getting it to work than you will getting future
application servers to work with the cloud. While I'm not a huge fan of
NFS I've been using it for many years with no problems due to changes in
NFS or operating systems or hardware. NFS has been available to the
public for about 20 years. Occasionally you may need to migrate it from
one platform or one machine to another but you very well need to do that
with clouds as well. Maybe you are using S3 but for whatever reason Sun
gives you a better deal with better terms and guarantees for using their
cloud. Maybe Amazon drops S3. Maybe because S3 moves servers to a
country that you are not legally allow to have your data in. Yes, you
have to plan for migration to new platforms but I fail to see how you
don't need to do that with the cloud. Really any major technological
decision should have an exit plan. Preservation storage is not different
in that and the cloud doesn't change that.
Edward
The power and cooling can be a savings, but in many cases the library or
individual departments don't pay for electricity, so while *someone* pays
the cost, it might not be the individual department. Cooling and electricity
costs are an actually a great argument for tape for large-scale storage.
Tape might seem old fashioned, but in many applications it by far offers the
best value of long term storage per GB.
It's true, tape is still an worthwhile option. Alternatives like optical or
magneto-optical media just have not kept up.
Again, I'm not totally against the cloud and there are some things I think
it could be very useful for, but the cloud doesn't make up for the lack of
(or just bad) planning.
Yeah, there's no system good enough to compensate for bad planning and
management.
--Joe
