[NTLUG:Discuss] Using SATA Drives to Replace Tape

[Chris Cox]

Robert Pearson wrote:
> On 1/9/06, Richard Geoffrion <ntlug at rain4us.net> wrote:
>>Robert Pearson wrote:
>>>Futures market bets big on start-up, 01/09/06
>>><http://www.networkworld.com/nlstoragealert16140>
>>>Chicago Mercantile Exchange IT leader finds futures market banks
>>>on storage start-up Copan.
>>Replacing tape backup with SATA drives.. Sure.. I do it and it works
>>DIRVISHLY.   I don't know about MAID though.  I don't personally know
>>any organization that needs that much backup storage.
>>
>>I like the spin-down technology though.  I can see how it can save
>>money.  Spinning = electricity + heat.  Heat = Electricity to cool.   I
>>am curious as to how they do spin up.  They HAVE to stagger the spin
>>ups... Can you imagine the power supply that would be needed to prevent
>>brownouts if all the drives came on at once!!!
> 
> The article claims growth from 4 TB to 180+ TB in 2.5 years.
> CDP looks like the only real solution.
> Veritas released the the CDP version of NetBackup in December 2005.
> The beta had been out since Spring 2005.
> At a "TB per hour" it would take 180+ hours for a "Full" backup.
> Not many sites have that kind of throughput capacity in the Infrastructure.

You can imagine how slow this is with tape at around 10-20MB/sec.

> 
> Quote from the article---
> "The duty cycle - when the disks are spinning - of Serial ATA drives
> is between 25% and 50%. The products from the other companies Taylor
> was looking at spun their Serial ATA disks 100% of the time, meaning
> that there would be more frequent drive failures."

We use 5 year warranted drives.  HOWEVER, NONE of the manufacturers
keep drive production for given series around long enough to
support 5 years.  So you DO need to keep a local inventory because
the replacement drives you receive from Segate-Maxtor, Hitachi or
Western Digital are NOT guaranteed to match one-for-one with the
model your are RMA'ing.  Something to keep in mind.

> 
> --- this sort of threw me. For rapid query response the disks need to
> be spinning. Staggered spin-up is probably the order of the day in these
> "giant" disk arrays. Anybody know? Selective spin-up is even better but
> I don't know anyone who knows the Content, or has it mapped, well
> enough to spin-up selectively. Unless Copan and MAID provide this.

It's ok... chances are their large Cisco switches make that power
draw look like a joke.  The mainframe likewise.

> 
> I was also under the impression that the key "Single Point of Failure
> (SPOF)" of a disk is the bearings. Thermally cycling the bearings has
> been the "Kiss of Death" for most disks. Spinning all the time was the
> work-a-round.
> 
> Just some thoughts...

We're running a large NFS shared home area off a 14 x 300GB storage
array and our disk-to-disk cache for backups is a 14 x 300GB array
plus a 14 x 400GB array.  The smaller units have been in production
for over 2 years now... we've only 3 drives... some of that was
our fault where one of those units was accidentally left on
during a mandatory weekend power outage (and therefore air conditioning
outage).

Our backups are rsync'd so you're never (hardly) copying the full
amount of data to the disk cache across the wire (we're now gigabit
end to end with a 10GigE back bone... with scattered 100mbit hosts).
Rsync is known as the destroyer and has been known for finding
a lot of problems in Linux that nobody noticed because folks weren't
driving their boxes very hard.  The advantage of using a continual
rsync is that your backups become pretty quick, unless all of
your data is constantly changing.

Pulling a drive array set to offsite is somewhat painful in that
the replacement array has to be initialized somehow.  It really
depends up on your backup schedule.  If it can take MORE than
a day to initialize, you're fine.  Odds are a new mirror won't
cause too many problems.  Now.. if you need this daily, then
it could be a problem.  We just do incrementals all of the time
to backup the machines (rsync) and then take the slow path
to the archive media (which is tape for us today... and
will likely be disk sometime this year).