[rescue] Backups

[velociraptor]

On 6/16/05, Jeffrey Nonken <jjn_rescue at nonken.net> wrote:
> On Thu, 16 Jun 2005 13:44:09 -0400, velociraptor wrote:
> > On 6/16/05, Jeffrey Nonken <jjn_rescue at nonken.net> wrote:
> >
> >> OK, good to know. I'm still not sold on DLT but I'll take a look.
> >>
> > Just a point of data: All linear tape solutions (DLT, AIT, LTO) are
>
> Um, AIT is 8mm helical scan. (A format which has never, ever given me
> a single problem, I might add. And the drives are self-cleaning and
> 100% duty cycle. :)

Sorry, mis-remembering...that was a while back.  Oddly enough
I was working at Sony (well "Sony Computer Entertainment
America", or, colloquially, the Playstation Division), and as there
was no internal channel for the aquisition of Sony hardware or
software, it was cheaper for me to purchase LTO.  Needless to
say, the engineers' desktops were Dells (the head end of the
IDE for the PS2 ran on Windows or Linux), and the artists' and
UNIX geeks' were Apples.

> QIC/Travan is also linear.

I didn't know that.

> > The most important consideration for maximum tape writing
> > efficiency and longevity with linear tapes/drives is the data
> > stream to the tape drive.  If there is not sufficient data being
> > fed to the drive, the tape will "see-saw" over the tape head
> > wearing both the tape and the tape head.
> >
> > The tape see-saws because data is written in bands onto the tape,
> > e.g. a hypothetical block looks like:
> >
> > ===
> > ===
> >
> As I understand it, the tape see-saws ("shoe polish") because the data
> are streamed and it has to splice the data on the tape. (It has to
> back up and get a running start so it's moving at recording speed when
> it reaches the end of the data.) One of the advantages of helical scan
> is that the medium is moving more slowly so it takes less time to
> reverse, and causes less wear on the drive.
>
> Unless you're talking about the fact that it (the linear methods I'm
> familiar with, that is) records to one end, then records to the other
> on alternate tracks (serpentine method), in which case I misunderstood
> what you were trying to say.

I read serpentine, as well, but maybe I misinterpreted the
way that it was used in the description of how the writes
are accomplished.  I interpreted "serpentine" across a block
only when not fed enough data.  I will have to do some more
reading.

An intial reskim of some of the info I looked at before indicated
that the table at the beginning of the tape is the major spot
for wear in all linear tapes, and suggests that the "1 million
passes" is in reality much (like <250K) actual "uses".  Of
course, you also have to look at the load/unload cycle life.

> > I suspect IDE will likely not be fast enough to feed the beast
> > unless you stream multiple jobs across the net- work.
> >
> I don't know about LTO. AFAIK most if not all AIT and DLT drives are
> wide SCSI, and DDS are various widths of SCSI.

I've never seen an LTO or DLT that was not wide SCSI, or
Diff/Wide SCSI.  I meant from the standpoint of streaming
data to them.  I'd seriously doubt the ability of an x86 box
with IDE drives to be able to provide the sustained output
to necessary to keep a local SCSI DLT streaming--but I'm
guessing.  Maybe SATA RAID-5?

=Nadine=