How to make 4.2BSD ``/etc/dump''s faster

[Chris Torek]

I've just succeeded in making our 750's tu80s stream when doing dumps.
This reduces the overall (elapsed) dump time by about a factor of 3.
That is, a dump that used to take 6 hours would now take about two.

There is some improvement on our 780's tu78, but not nearly as much;
it seems to cut off about 1/3 of the time (3 hours => 2 hours).

If there is enough interest, I will (eventually) post the changes and
new sources to net.sources and/or net.unix-wizards.  It has the
following properties (some good, some bad), so you are forewarned.

	- It works with just about any tape drive, though you'll need
	  to make changes to the driver code.  I have changes for
	  vaxuba/ts.c, vaxmba/ht.c, vaxmba/mt.c, and vaxmba/mba.c.

	- It has not been extensively tested and debugged.  However, it
	  isn't too terribly complex, so there is a low probability of
	  remaining bugs.

	- It requires changes to some of the existing kernel code.
	  As kernel changes go, they're fairly trivial, but they are of
	  course sensitive.  As we are not running a "stock" 4.2
	  kernel, there will undoubtedly be incompatibilities between
	  our kernel and your kernel.  In other words, it won't fit in
	  without some effort.

	- It is not complete.  You can only write(), not read().  (I
	  haven't needed read()....)

	- It requires changes to the dump program.  These are mostly
	  pretty minor, just the addition of new routines.  (Right now
	  I have a separate program, but I plan to merge the code into
	  the regular "dump" and make it an option, somehow.)  (Of
	  course, I've made lots of other small changes inside dump
	  too...)

It should be usable for things other than dumps; however, this is
the only thing that we do regularly which also consumes lots of
machine time (since every other week the Vax is down for hours
doing level zero dumps), so this is all I've done with it.

If you want to see this posted, send mail.
-- 
(This line accidently left nonblank.)

In-Real-Life: Chris Torek, Univ of MD Comp Sci Dept (301) 454-7690
UUCP:	{seismo,allegra,brl-bmd}!umcp-cs!chris
CSNet:	chris at umcp-cs		ARPA:	chris at maryland