[101] in Info-AFS_Redistribution
re: fms & DAT drives
daemon@ATHENA.MIT.EDU (Pervaze_Akhtar@transarc.com)
Tue Apr 23 13:40:51 1991
Date: Tue, 23 Apr 1991 13:22:13 -0400 (EDT)
From: Pervaze_Akhtar@transarc.com
To: Info-AFS@transarc.com
Cc:
We've not had the opportunity to run fms on a DAT drive. I doubt if we'll
be in a position to do so in the near future. Sharing your results via the
Bboard would be useful.
What does fms do?
-----------------
It is an almost trivial program. Pass 1 fills the tape with 16k blocks.
Pass 2 writes alternating 16k blocks and filemarks. Upon completion it
does a simple computation to determine the size of file marks. The "progress"
displayed on the screen is updated periodically, but not after every block.
It could be after every 2 blocks [I forget the value used].
Expected performance
--------------------
On all drives that I've tested, the first pass goes fast. In the majority
of cases, the data is supplied to the drive fast enough, that "streaming"
takes place, i.e. the drive does not have to stop the tape between each
write.
In pass 2, the writing of file marks results in non-streaming mode, in most
drives. The result is that after every operation (or pair of operations),
the drive stops the tape, backs up (i.e. repositions), and only then will
it accept the next write. This results in operation that is often an
order of magnitude slower.
DAT drives
----------
My understanding (from literature) of DAT drives, is that they have a capacity
of approx. 1.3G and use "encoded" file marks (as opposed to file marks that
are a block of blank tape).
From the first pass of 79765 blocks, the 1.3 G capacity seems correct.
The behaviour of pass 2 suggests either an extremely inefficient implementation
of file marks, or use of "blank tape" filemarks.
It may be that fms should be smarter and use less filemarks in its capacity
evaluation, since some drives incur a much larger performance penalty than
we anticipated.
