HP3000-L Archives

April 1999, Week 2

HP3000-L@RAVEN.UTC.EDU

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Carl McNamee <[log in to unmask]>
Reply To:
Carl McNamee <[log in to unmask]>
Date:
Tue, 13 Apr 1999 07:41:31 -0500
Content-Type:
text/plain
Parts/Attachments:
text/plain (70 lines)
We performed some testing last night by backing up to device $null and found
that our test data was I/O bound.  That would explain why we never saw an
increase in speed for the DLT7000 over the DLT4000.

Thanks to everyone for your input!

Carl

-----Original Message-----
From: Jeff Woods [mailto:[log in to unmask]]
Sent: Monday, April 12, 1999 2:16 PM
To: Carl McNamee
Cc: [log in to unmask]
Subject: Re: [HP3000-L] DLT7000 vs DLT4000?


Hello, Carl.

At 4/12/99 01:23 PM -0500, Carl McNamee wrote:
>We have been using dlt4000 tape drives for a couple of years now and wanted
>to shorten our backup window so we decided to try the dlt7000.  When I
>finally received one and installed it on the system with all the latest DLT
>and Turbo Store patches we were surprised to find the backup times were
>exactly the same as on our DLT4000's!
>
>Puzzled by this I had several tests run; with compression, without
>compression, ect...... and each time the backup took almost exactly 30
>minutes.  The dlt7000 is connected to its own f/w scsi controller and is
the
>only device in its card cage off of our 997/500.
>
>Has anyone on the list had a similar experience?  Any thoughts on why the
>tape drive would be so slow?

Because the tape drives are not the limiting factor in your backup speed.
It's
probably time spent fetching data from disk and preparation for sending to
tape.  Are you interleaving files on tape?  I know that some folks really
don't
like that (Hi, Stan! ;) but the purpose for interleaving is that when
multiple
files are being written to tape at the same time, the chances that those
files
may be on multiple disk spindles goes up significantly, and thus the time to
read all the files goes down due to the more parallel nature of the disk I/O
that permits.  That's why RoadRunner by default uses an interleave of 4 and
allows the user to make that as high as 8 files concurrently open for
storing
per tape drive.

Also, RoadRunner doesn't read a file which spans multiple drives in logical
sequence.  It is aware of which extents fall on which spindle and reads
extents
on multiple drives concurrently as much as possible.  This also helps
increase
the parallelism of disc reads during the backup.  I don't know whether
TurboStore also does that, but with the TurboStore default of no
interleaving,
it's my understanding that the files are always stored in logical order on
tape...  hence the file has to be read one extent at a time in order.

If you use RoadRunner, which if I recall correctly you used to use there,
try
increasing the interleave to 8.  If you use TurboStore, then add the ;INTER=
option and run your test again.  I would be very interested in seeing the
results.
--
Jeff Woods
[log in to unmask]  [PGP key available here via finger]

ATOM RSS1 RSS2