On 01/23/2012 02:46 AM, R.S. wrote:
W dniu 2012-01-22 05:21, Joel C. Ewing pisze:
I've said it before, but will say it again: modern tape media has such a
large capacity that a single dfhsm cartridge can contain an incredibly
large number of datasets. It is almost inevitable that loss of a single
dfhsm ML2 or Backup cartridge will impact something you care about or
can't afford to lose (or are legally required to retain), and it is also
inevitable that single cartridges will occasionally be physically
damaged. Over the long haul, you really can't afford NOT to duplex all
dfhsm carts (and for that matter, non-dfhsm carts that contain data you
can't afford to lose).

90% Agreed, some remarks:
1. Cart capacity is not so relevant. The problem exists with any cart
capacity, the difference is in a likehood, but the likehood is quite
similar (same order of magnitude).
2. It worth to separate two cases:
a) Backup copy loss. It's more or less like loss of spare in a RAID
group, you did not lose your data, you lost redundancy. Additional event
must happen to make backup copy needed for recovery. Of course there
could be some legal requirement to have such copy, in such case he
requirement is broken.
b) Archive copy, ML2 (without backup). In such case lost of cartridge
simply means lost of data.

I always vote fo duplexing data on tapes, for the same reasons why we
use RAID protection on DASD. I mean real media, for example virtual
tapes are usually on RAID-protected disks.


At least based on our experience, I would disagree on point (1). Over the long haul (after drives and cartridges begin to age) we saw about the same unreadable cartridge rate (inability to later read a cartridge that was written successfully) among 3490 and 3590 - more than one and under five losses per year (typically physical damage to cart or loss of data on cart from some drive failure). The difference in capacity (from 3490) was significantly more than an order of magnitude, so I believe our risk of significant data loss also went up.

The newer tape technology tends to be more reliable, with fewer physical volumes and fewer physical mounts, but this can also be offset by a much larger number of physical mounts and opportunities for physical risk for the fewer media volumes that do contain active data. Unquestionably, the potential data loss per cartridge loss (and likelihood of visibility of a single incident to management and end users) is directly proportional to the data capacity of a cartridge.

Tape media capacity over the last several decades has increased by over 3 orders of magnitude. Any installation that may have dismissed duplexing when on 3480 or earlier tape technology and hasn't since re-evaluated that decision should do so. Today no one would (I hope) consider running z/OS production on DASD that wasn't covered for single media failure by RAID or mirroring. Similar respect is needed today for data retained on real tape media.

--
Joel C. Ewing,    Bentonville, AR       jcew...@acm.org 

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN

Reply via email to