[nylug-talk] Recovery is not about media, but availability and storage -- WAS: backup media

[Bryan J. Smith]

Alex Pilosov wrote:  
> Another thing I wanted to bring up:
> Backups on tape may be soon outdated, and we'll think of them as quaint
> as punchcards.  The current HDD pricing is comparing very favorably with
> cost of tape media.

Cost isn't typically why tape is still used as the preferred 5-10 year
"off-line" medium.  Commoditization of disc has helped in the "cost"
versus tape, but tape still "wins" on surface area, let alone "off-line"
longevity and storage.

Although 2.5" disc is gaining in popularity for shorter-term 2-3 year
"off-line" usage.  E.g., some newer tape cartridge solutions are
actually just commodity 2.5" disks packaged.  2.5" disks can handle
almost as many Gs as a typical tape cartridge.

Alex Pilosov wrote:  
> For example, 500GB HD is ~100$ (20c/GB).

A commodity, 3.5" HD is designed to be used on-line at least a few times
a month, typically once a week at least.  Failure to spin a 3.5" disk
can cause its low-cost, commodity materials to break-down.  E.g., Even
most near-line backup solutions "spin" and "exercise" commodity, 3.5"
HDs several hours at least once per week -- typically doing a checksum
verification.

Alex Pilosov wrote:  
> However, random access drive is much easier to deal with - no delays,
> near-0 cost of reader, which results in ability to pack *lots* of
> backup storage in small space (vs huge tape changers).

Which is why disk started replacing tape as the preferred "near-line"
medium about 5 years ago.  Basically around the same time near-line
solutions hit 100TiB, tape was overtaken by disk solutions.

But tape is still the primary "off-line" medium, even today.

Alex Pilosov wrote:  
> I think we will see a very-very-low-cost huge HDD "storage" arrays
> (no full access to all drives concurrently, instead, a backplane with
> software-selectable access). I don't know of a product like that yet,
> but I'm sure there'll be one soon (think low-cost "thumper"-like
> chassis)

You mean the concept of the "Virtual Tape Library" (VTL) and the
evolution of disk concentration into the "Massive Array of Idle
Disks" (MAID)?

The "Virtual Tape Library" (VTL) that has been around for almost the the
last decade.  FalconStor VTL is probably the most proliferated solution,
originally released on Red Hat Linux 7.3.

'MAID" was coined by several people, but the person who put it to work
was Dr. Aolke Guha, former VP and Chief Architect at StorageTek (a major
tape archive/library product vendor, now owned by Sun).  Right after his
2002 January USENIX presentation, he founded Copan Systems -- the
company (last time I checked) -- with the greatest disc concentration
I've ever seen (896 drives in 32U).

RE: "Dissecting Virtual Tape Libraries", Sys Admin 2005 September  
  http://www.samag.com/documents/sam0509a/  

Even Microsoft has had that exact solution in-house for almost the past
5 years.  ;)

For those that can't afford a VTL solution, I hacked out a quick blog
entitled "Common Sense Disaster Recovery":  
  http://thebs413.blogspot.com/2007/03/common-sense-disaster-recovery.html  

There are basic fundamentals to disaster recovery. They are:

  1. The 3 basic levels of data redundancy
    A. On-line: immediate recovery of data on the system itself
    B. Near-line: near-immediate recovery from another system on the
network
    C. Off-line: disaster recovery when you don't have the system, let
alone the network

  2. The time crunch, especially recovery time
    A. Consistency/Reproducibility
    B. Boot to Restore
    C. Network Reconfiguration

  3. Test restores are an ongoing duty
    A. On-line: Check options daily
    B. Near-line: Daily logs, check restore before you off-line
    C. Off-line: As you off-line, as well as when you rotate, from
near-line

Again, disaster recovery is about recovery, not backup.  If you only
have short-term, on-line or near-line solutions (among other, missing
pieces), you're incomplete.

Alex Pilosov wrote:  
> Historically, while cost per byte for HDDs has been dropping fast,
> cost of per byte for magnetic tape media has been dropping far slower.
> I think in a year or two, we'll see HDDs catch up with cost per byte.

Again, cost isn't the reason why tape is still more popular for an
"off-line" medium.  Off-line longevity of 3.5" disc absolutely sucks,
especially when it comes to taking Gs and leaving off-line for weeks.
Commodity 3.5" disks aren't designed to be spun-down for long-periods.
That's why many of them arrive DOA when they've been off-line for
months.

Although as 2.5" disc becomes more and more commodity, especially as its
quickly becoming the defacto-standard disk form-factor in many data
centers, the prices will make it a compelling alternative for even
off-line.  With its increased off-line longevity and high tolerance to
Gs, at least compared to 3.5", 2.5" is a viable, off-line solution for
2-3 years.


Ron Guerin wrote:   
> I've already taken all my customers off tape, external drive backup
> is where it's at.

Using what, external connection?

One of the major reasons I saw for downtime in various Boeing labs
(e.g., confidential and classified labs that were not on centralized
networks) was external FireWire and/or USB devices attached to servers
and 24x7 workstations.  Once I yanked them, that removed the problem.

An inexpensive, dedicated system with near-line 3.5" disc storage you
rsync to nightly, and then put to an off-line tape or 2.5" disc medium
every week or so, is the most ideal solution I've deployed when cost is
an issue.

Ron Guerin wrote:   
> It also removes the most likely point of failure in tape 
> backups, that is the people involved in changing tapes.

So what do you use for an "off-line" medium?  What if your near-line is
destroyed physically or erased on the network?

On-line, near-line and off-line -- you're either complete or incomplete.


Joshua Zeidner wrote:  
> Holographic storage is another cool thing in the pipeline:
> http://en.wikipedia.org/wiki/Holographic_memory
> the capacity of holographic media is many orders of maginitude
> greater than 2-d media like CD-ROM or HDDs.

What is most ideal about crystal-laser (CL) and other types of
holo-storage is the longevity, assuming the kinks get worked out.  That
is why it is most likely to replace tape.

Magnetic disk and EEPROM (flash) will not replace tape.  It's a simple
matter of surface area in the case of the former, cost and size in the
case of the latter.  But CL is now about volume, not just surface area.

Hence your quite probable prediction, come 2015 or so.


-- 
Bryan J  Smith              Professional, Technical Annoyance
mailto:b.j.smith at ieee.org  http://www.linkedin.com/in/bjsmith
-------------------------------------------------------------
           Fission Power:  An Inconvenient Solution