In <bitnet.adsm-l%_12232_Fri_Sep__1_11:37:[email protected]>, pshields AT bnr DOT 
ca (paul p. shields) writes:
>I am looking for people who have adsm configured in an environment similiar
>to ours. We are currently looking at backing up about 1.4T of data spread
>across about 200 machines.
>
>The important part is the following. Our average file size is in the 20 -
>40Kb range which under our current evaluation and test environment seems to
>be stressing the database on the RS/6000-based server.
>
>Is there anyone out there with similiar average filesizes, backing up at
>least several Gigs of data that would be willing to iscuss your experiences
>in performance and configuring the servers for optimal performance.
>
>Paul Shields

I am within an order of magnitude of your system, but on an MVS server.
Version 2 of the server and clients contain parameters which can tune
how often a commit is done in the database (I am guessing that this is
the cause of the 'stress'.)  Without the parameters the default is to
commit whenever the server gets 10 files in a transaction, or whenever
the client has sent 300k.  With the server parameter TXNGROUPMAX you
can raise it to 256.  The client parameter TXNBytelimit can raise it
to 25600k before the client forces a commit.

I tested these on a mac along with increasing the tcpbuffersize and
windowsize and got double the performance.

What exactly is the nature of the database stress?

Bill Colwell
C. S. Draper Lab
Email: BColwell AT draper DOT com
Voice: 617-258-1550