>... If I
>could increase the buffer size to 1199840 then I could achive about 15
>megabytes a second. As it is now I am wasting space and time writing 32K
>blocks.
1199840 is a strange number (as in, not a power of 2). Could you live
with a blocksize of 1199104 (1171 KBytes)? That would waste 736 bytes
on each track (less than 1%). It might be interesting to try some dd
tests with 32 KBytes and 1171 KBytes to see how it does.
Is 1199840 usable data, or does it include stuff the drive adds?
In any case, I can give you a larger blocksize right now, with some
caveats. Go to the FAQ and look for the couple of articles on how to
build Amanda from the CVS sources. You want branch amanda-242-tapeio.
Then just set blocksize in your tapetype as needed (see the amanda(8)
man page).
Note that this code is brand new. I *think* it's right, but other than
the testing I did, it is not in production anyplace.
>In the current version of amanda is there any way to pipe the output thru
>"dd" to change the block size?
It would be pretty hard. Taper is a complicated I/O processor built for
speed and it also needs to be able to deal with error handling. It's not
a simple Unix process in a pipeline you can tack something on the end of.
Such is the price of progress :-).
You might be able to increase TAPE_BLOCK_SIZE in amanda.h (my concern
before was when you wanted to decrease it). I think one or two other
people have done this, but I don't know what else they had to change to
get it to work.
>Robert Minsk
John R. Jackson, Technical Software Specialist, [EMAIL PROTECTED]
