[rescue] Block size and the single DD

[Sheldon T. Hall]

So I've got this Solaris shell script that backs up my SGI machines to the
spiffy new DLT tape drive on my SPARCclassic....

The heart of the script is this line:

	rsh $MACHINE xfsdump -l 0 -F - $FS | dd of=$TAPE bs=$BS

I run that on the Sun with values like

	MACHINE=blinky	# Indigo2 R4k
	FS=/dev/root
	TAPE=/dev/rmt/1cn	# DLT 2000xt on Sun SBUS card
	BS=1024k

Now, this works, but it's so slow that the DLT tape shoeshines itself into a
lather.

I've tried values for BS from 128b to 1024k, and none solve the problem.  At
the low end of the BS scale, the tape flips back and forth constantly, while
at BS=1024k, there's a slight pause between each tape movement.  Nothing
much is running on either box, and the network (10baseT) is not busy.  I've
moved the DLT from the SPARCclassic's built-in 50-pin SCSI to a 68-pin
Sunswift SBUS card, too, without much effect.

The next thing I'm going to try is getting the 100baseT running on both
machines, but the blinkenlights on the various network pieces make it look
like it's not filling the pipe it's got....

I run the script as a "pull" backup from the Sun because it's always up, and
the SGIs aren't.  The Sun backs itself up to the DLT with a different
script.

Have I left out some important factor that would make this hum?

Should I put the tape drive on one of the SGIs (I have a Challenge L and the
I^2) and do it the other way around?  Use rmt from the Sun to the DLT on the
SGI?

Thanks.

-Shel