Subject: | |
From: | |
Reply To: | |
Date: | Mon, 22 May 2000 09:19:47 -0700 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
Carl writes:
>My experience has been, on systems ranging from a 930/020 to a 997/800, that
>MPE cannot take full advantage of any tape device faster than a DLT4000.
>There is some limitation within MPE/hardware that bottlenecks the i/o at
>around 5 Meg per second. You can test this theory on your system by doing a
>backup to $NULL and then checking the through put portion of the statistics
>listed at the end of the backup.
<semi plug alert>
I need to clarify this a bit ... MPE is capable of better performance, but that
is based upon a number of factors such as configuration, mix of devices on the
SCSI bus, number of bus converters, etc. The problem you're seeing is most
likely not an MPE problem per se:
Here is a store to $NULL of a 51gb file using Backup+. This system is a 989/200:
>:file z=$null
>:listf large01.prodgp01.abstesta,11
ACCOUNT= ABSTESTA GROUP= PRODGP01
Name Access FCode RecSiz Type EOF File Limit Disk Usage Exts
-------- ERWS ----- ------ ----- ------------- ------------- --------KB -----
LARGE01 256 FB 210000000 210000000 52500000 *****
>store large01.prodgp01.abstesta;*z
Building intermediate scratch files ...
Selecting files for store ...
1 files scanned ...
1 files - 210000000 sectors (50.07 gigabyte(s)) of disc space to store
Optimizing selected file information for faster backup ...
Storing selected files ...
Store is 100% complete
..
total amount of disk space stored: 210000000 sectors (50.07 gigabyte(s))
total number of tape errors: 0 + 0 errors in store directory
total number of tape retries: 0 + 0 retries in store directory
This store took 18 minutes, 10 seconds
Note this took 18 minutes as a store to $NULL. That indicates an aggregate
throughput of more than 5MB second, so the problem isn't MPE.
We can store the same file to a DLT8000 in two hours. Also note that our
configuration is not optimized for performance ... we have a number of 36gb
disks used for our test environment and this doesn't allow the benefits
of overlapped IO that would happen when using a larger number of smaller
disks.
BTW - we did a performance test recently that can be independently
confirmed where we were able to achieve a throughput of over 1gb
per minute to a single DLT8000. That machine was a 997/12 way with
30+ spindles of 4gb each, spread over two bus converters. The DLT8000
was the only device on it's SCSI channel.
As always, YMMV.
|
|
|