Subject: Re: Problem with Compaq DLT streamer
To: Greywolf <greywolf@starwolf.com>
From: Brett Lymn <blymn@baesystems.com.au>
List: port-sparc
Date: 08/17/2004 10:52:11
On Mon, Aug 16, 2004 at 09:22:08AM -0700, Greywolf wrote:
> 
> "Tape Blocking" is merely a term assigned to writing data of a fixed
> length -- whether or not it increases performance depends upon the blocking
> factor.
> 

Which I did say :) you max the performance when the tape is streaming.

> 
> Are you sure it's 512 across the board? 

It looks to be machine dependent but I believe that all NetBSD
architectures use the same size.  Have a look for DEV_BSIZE in
/usr/include/machine/param.h


> I thought it was 2k, or, at
> the very least, 1k, which is the smallest blocksize some tape drives
> are known to handle.  I know, for example, that EXB-8x0y tapes didn't
> handle anything less than either 1k or 2k (512 caused a driver error).
> 

There you go - you got the block driver block size confused with tape
blocking.... they are not the same thing.

> 
> Pity.  I know it seems archaic, but as tape will really never go away,
> realistically speaking -- at least, certainly not in the near future --
> it would be very cool to support tape booting.
> 

And very slow.... I have some non-fond memories of installing SunOS
4.1.0 from a QIC tape during a training course.  They timed the course
so we did all the prep just before lunch and went to lunch while the
OS was being installed.

> By the way:  I never EVER knew of a re-writable FS on tape, not even with
> individual blocks, simply because of the geometric unpredictability of
> tapes

As others have already said, VAXen were capable of doing this - I know
that the old 9 track tape subsystems were capable of doing block
rewrite and positioning.

> 
> Now all I need is to get my 850x up and going (and I should see if I can
> get the changer and the drive to interact), and maybe I could actually
> get a fs-on-tape (1k/1k)...
> 

That may be _very_ suboptimal, exabyte tape drives really do suck at
positioning and writing small amounts of data, they are designed more
for streaming backups to :) also, I am reasonably sure that exabytes
write a double EOT marker when the session is closed which would
probably cause grief.

-- 
Brett Lymn