On Tue, 27 Apr 2004 at 12:00pm, pll+amanda AT permabit DOT com wrote
> What variables affect "Run time" as reported by amanda after a run
> is complete? For example, in my reports, I see the following:
>
> Estimate Time (hrs:min) 2:49
> Run Time (hrs:min) 10:33
> Dump Time (hrs:min) 8:38 8:38 0:00
>
> So, it took almost 9 hours to dump to tape, but 10.5 hours of actual
> run time. Does that imply that the first 2 hours were spent getting
> estimates and then spooling enough data to the holding disk before
> streaming to tape began (i.e. is the 8:38 in parallel with the 10:33) ?
You're a bit off in what the numbers mean:
Run Time is the time from which you start 'amdump' until everything is
done and amanda sends out the report mail.
Estimate Time is what you would expect (how long it takes amanda to do all
the estimates -- no bits are actually coming off the spindles during this
time).
Dump Time is how much time amanda spends actually getting bits off the
disks. They go either to the holding disk, or straight to tape.
Tape Time (which you left off) is how much time is spent actually writing
to the tape.
That your Run Time > Dump Time indicates that either you have a very low
number of clients and/or not enough holding space. In either case, you're
not getting much benefit from amanda's parallelism. I'm guessing some
more holding space would help you quite a bit.
Also, your estimate time is quite long. Working on that is a
hardware/OS/FS issue, not really an amanda one. E.g., I see long
estimates using tar on an XFS on Linux partition with *lots* of *very
small* files (which XFS isn't all that good at). In my case, I got the
user to pack up those files, and saw estimate times go down.
You can look in the amdump.N files on the server to see which client(s)
is/are taking the longest to do estimates. Then look at the
/tmp/amanda/sendsize*debug files on the clients for the command you need
to test and optimize.
> Currently I'm using hardware compression on an LTO2 drive. Presumably
> things might be faster if I turned off hw compression and used the
> clients to do compression before sending to the amanda server?
Maybe, maybe not. It depends on the client CPU speeds. Also, if I
understand correctly, LTO2 drives are smart enough to not compress already
compressed data, so you may be able to "get the best of both worlds", and
use a mixture of hardware and software compression[1].
So, to sum up, add some more holding space, look into speeding up your
estimates by optimizing your disk systems, and maybe use some software
compression here and there. Good luck.
[1] NOTE: On almost all tape drives, this is a Very Bad Idea. Trying to
hardware compress data already compressed in software generally
expands said data. But, again, I think LTO2 drives are smart enough
to only compress when it'll actually help.
--
Joshua Baker-LePain
Department of Biomedical Engineering
Duke University
|
