On Wed, Dec 01, 2004 at 08:57:05AM +0100, Nicki Messerschmidt wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Hi list,
> I have a very strange phenomenon. There is a server with an Seagate
> Scirroco 12/24Gb DDS3 drive attached. Onto this drive I can put data via
> 'dd if=/dev/zero of=/dev/nst0 bs=1024 count=100000' (100Mb) in about
> thirty seconds. This means a transfer rate of about 3.3Mb/s.
> But if I do an amflush of 6547621k it takes forever (about 12h).
>
> What is wrong?
>
> Here the relevant parts of the amanda.conf:
> taperalgo smallest
> dumporder "SSSSTTTTst"
> tapedev "/dev/nst0"
> tapetype HP-DAT
> define tapetype HP-DAT {
> comment "DAT tape drives"
> length 11300 mbytes # these numbers are not accurate
> filemark 512 bytes # but you get the idea
> speed 4500 kbytes
> }
>
> How can I debug this strange thing?
I question the validity of your test. /dev/zero as a data source
totally eliminates any disk activity involvement in your test.
Also eliminated is any bus contention for accessing the data to
memory then transfering it, possibly on the same bus to the tape.
And if the drive is in hardware compression mode, /dev/zero provides
a wonderfully compressible source of data. Your 100MB may have
taken about 1 inch of tape :))
I have never noted any DDS3 tape giving much over 1MB/sec write rates.
Even so, at that rate a 12GB tape should be writable in about 200 min.
I'd suggest you create a large file, certainly over 1GB, on the same
filesystem as your holding disk. The do your dd test. Don't create
the file from /dev/zero. /dev/random would be a reasonable choice.
Or I somtimes 'cat' collection of different types of files into a
temp file. This cat then be repeatedly cat'ted >> test_huge_file
until you get a large enough file. Do the dd with a bs=32k to
match amanda's blocksize and be certain the tape is in the same
compression setting as amanda uses.
Was your amflush a single or repeated observation. If done during
a period of high activity on the system other activity might have
prevented the tape from being fed at a rate which allows streaming.
If it is not streaming you would probably hear it "shoe-shining"
from all the forward/reverse/forward movement of the tape. That
is a major cause of slow writes.
jl
--
Jon H. LaBadie jon AT jgcomp DOT com
JG Computing
4455 Province Line Road (609) 252-0159
Princeton, NJ 08540-4322 (609) 683-7220 (fax)
|
