ADSM-L

Re: Performance issues and stats

1996-08-02 16:05:40
Subject: Re: Performance issues and stats
From: "Pittson, Timothy ,Corp,US" <tpittson AT HIMAIL.HCC DOT COM>
Date: Fri, 2 Aug 1996 16:05:40 -0400
Jon,
        Don't know much about FDDI but here are a couple of things to consider
on the Netware side of things..

1) If you're running Netware 4.1 and using the file compression feature,
do not use ADSM compression - too many retransmissions can slow things
down a lot.

2) On incremental backups, I've found that the number of files on the
client can be a better indicator on how long the incremental will take -
ADSM needs to check its database to see if a file has been changed.
Some of our Netware servers have 400,000 + files on them and can take
forever for an incremental backup to finish, even if you're only backing
up a relatively few files and little data (i.e. 2,000 files, 300 MB).
Check out using the INCREMENTALBYDATE option ('-incrbydate') on some of
your Netware incremental backups.

3) TCP/IP on Netware - check out the tcpmssinternetlimit option for your
TCPIP configuration on Netware.  By default, when going to remote
networks, Netware will break things down into small packets (576 bytes).
 By setting tcpmssinternetlimit to off, Netware will send bigger packets
I've found that this can help quite a bit.

4) TXNBYTELIMIT may or may not help you out - I've found it helps us out
most on our Netware servers that act as MSMAIL servers.  The average
file size on these servers is pretty big and the data is very volatile
(i.e. over 75% of the data on the server gets backed up every night).
By default, I set it to 8192 for most of our Netware servers and higher
(16000 - 25600) on our MSMAIL Netware servers.

5) TXNGROUPMAX goes hand in hand with TXNBYTELIMIT.  Only problem with
setting this too high is if you have compression enabled and one of the
last files being transmitted 'grows' when compressed. the whole group of
files is transmitted.  Again, this was a big problem for our Netware 4.1
servers until we turned ADSM compression off for them.  I've found
having it set to 64 works well for us.

Good luck !!!
Tim Pittson
tpittson AT himail.hcc DOT com

>----------

>From:  Jon C. Austin[SMTP:TSSJCA AT SYSA.ADM.DUKE DOT EDU]
>Sent:  Friday, August 02, 1996 3:14 PM
>To:    Multiple recipients of list ADSM-L
>Subject:       Performance issues and stats
>
>We're on the edge of upgrading to V2 Server and possible putting a
>OSA1 FDDI in our S/390 MVS where our server resides. Part of our just-
>ification for this has been to perform incrementals on more clients
>during an off-hours of window of 9p-4a weekdays. Now we're trying
>to make sure we haven't got anything out of tune anywhere, on clients
>and on the server before we do this ( to prevent the egg on face
>syndrome). Let me provide this topology:
>
>   Server: 9672 R22 S/390 MVS 4.3.2
>           DISK: RAMAC1 escon attached
>               Disk pool spans 6 volumes, and database on a separate
>vol
>           Tape: 3494 Library escon attached all 3490 drives
>           ADSM server level v1r2m14
>
>   Network: all TCP/IP traffic
>              RS/6000 Channel-attached
>              Token-Ring in RS/6000
>            campus topology of FDDI Ring and Cicso Routers, so most
>            (not all) backups flow:
>
>      client--->cicso--->FDDI---->cicso---->RS/6000---->Server
>
>    most clients on 10MB Ethernet, some on 16 MB Token ring.
>
>
>   Clients: mostly Netware/Solaris/AIX/OS/2 in that order.
>
>
>
>  About 20 clients kick off 9p-9:15p (half netware/other half unix)
>
>  Most UNIX clients are done in 45min-2 hours depending on incremental
>  need.
>
>  Other groups of Netwares come into the mix at 10p 11p and midnite.
>    No more than 3 at a time.
>
>
>  Most netwares finish, but can ususally take 2x or 3x time to back
>  the same amount of data as the UNIX clients (most netwares are
>  486 or higher CPU, 3.x and 4.x). Some will not even finish and
>  only get around 350MB through in 7 hours!
>
>
>I've been searching IBMLINK for tricks. Can anyone suggest some more?
>
>I'll let you know, most of the netwares have their TCP buffers and
>windows maxxed. I haven't messed much with the TPXGROUPMAX and
>TXNBYTELIMITs, but would appreciate any good or bad info about tweaking
>these parameters.
>
>Much appreciated.....
>
>/--------------------------------------------------------------------\
>|  Jon Austin                             TSSJCA AT SYSA.ADM.DUKE DOT EDU   |
>|  Systems Programmer,                                               |
>|  OIT Systems Administration Group                                  |
>|  412 North Building, Box 90132                                     |
>|  Duke University                           phone   (919)660-7024   |
>|  Durham, NC  27708-0132                    fax     (919)660-7029   |
>|               Duke University: Birthplace of 'USENET'              |
>\--------------------------------------------------------------------/
>
<Prev in Thread] Current Thread [Next in Thread>