3581 Tape Drive performance
tata...@yahoo.com
we have a 3581 tape library with 3580 tape drive in it.
IBM Specs says that this tape drive give up speed up to
35MB/Sec.
Block size is zero. I copied 21178MB data on tape, it took 30 minutes
and 56 seconds (11.41MB/sec) as per specification the speed should be
35MB/sec. We are getting 32% performance of the tape.
How we can improve the performance of Tape Drive
What are performance limiting factors in this regard....???
Bogus Strawman
I've not used this particular device but in general you need to be able to
feed the tape drive fast enough for it to keep streaming (writing),
otherwise it has to stop because there is nothing to write, rewind a bit,
then start again. You don't say what machine you have this on - the
machine's bus and drives need to put out that kind of sustained data rate,
it also helps to put the drive on its own scsi bus.
cheers
tata...@yahoo.com
It is on a p650 Server and attached to scsi2 adapter ..... As far as
speed of the bus or scsi adapter is concerned how can i know that???
What type of test can i run to see how my filesystems are performing //
that is , from where the data being read....
Greg Beeker
as I understand, you only want to use block size of 0 (which means
variable) for compatibility to read with existing data. You may want
to experiment with larger block sizes to increase performance.
Hope this helps!
dfuninpa
zero will yeild the most performance for those particular drives.
More than likely you will never be able to utilize the entire bandwith
of the drive without some advanced mulit-stream capable backup software
such as Legato Networker or Tivoli Storage Manager.
Hope this helps.
-D
tata...@yahoo.com
As far as i understand from u guys responses is that it is not possible
to achieve documented speed of these tape drives ....... if only with
any specific software , these tape drives can acheive documented speed
then documents should also identify these softwares or that scenerio...
what u guys think about that ?????
Jose Pina Coelho
@f14g2000cwb.googlegroups.com:
Wrong. Most software can do it if properly buffered.
Try dd if=/dev/zero of=/dev/rmt0 ibs=1m obs=32k
Laurenz Albe
>> As far as i understand from u guys responses is that it is not possible
>> to achieve documented speed of these tape drives .......
>
Yeah, but wrong only if your software reads from /dev/zero.
In real life, data that are written to tape are usually read from
somewhere, frequently a hard disk.
Buffers don't help you much with large sequential reads.
No matter how big your buffers, they will run full.
And to fill them in the first place, you must read the data from the disk.
Yours,
Laurenz Albe
Bernardo Cabral
>
> Yeah, but wrong only if your software reads from /dev/zero.
> In real life, data that are written to tape are usually read from
> somewhere, frequently a hard disk.
>
> Buffers don't help you much with large sequential reads.
> No matter how big your buffers, they will run full.
> And to fill them in the first place, you must read the data from the disk.
And that's not a tape speed problem.... the tape can still make the
announced speed, but the other component can keep up!
tata...@yahoo.com
In last reply from Bernardo , it says " it is not a tape speed
problem"... how one can assure on this thing!!!. Either it is tape
drive itself or filesystem which contains data
Guys ! Do you have some test to distinguish between these two????
Laurenz Albe
> Either it is tape
> drive itself or filesystem which contains data
> Guys ! Do you have some test to distinguish between these two????
Maybe use the test Mr. Coelho suggested:
time dd if=/dev/zero of=/dev/rmt0 bs=... count=...
to see how fast you can write to the tape (of course in this case
the bottleneck could also be the bus).
time dd if=/dev/hdisk1 of=/dev/null bs=... count=...
or something similar to see how fast you can read from disk.
Don't expect any device to perform as fast as the specification
suggests :^)
Yours,
Laurenz Albe