ADSM-L

[no subject]

2015-10-04 17:47:36
From: Maria Jose Rodriguez Canales <mar_jose AT ES.IBM DOT COM>
To: <ADSM-L AT VM.MARIST DOT EDU>
Tjeerd, I'll try to give you all of the information I've got. The ADSM V2
server and client are running on the same machine, a Netffinity running OS/2
Warp Server. The communication protocol is Shared Memory.

  We first tried backing up all of the data to a VSE disk accessed through
LANRES. The results were really poor because to back up about 350 MB it
spent
about five hours or so. The communication between OS/2 and VSE is
complicated,
with a lot of network such as 3745 communication controllers and so.

  The first time we tested backup with this kind of configuration, we only
used
the default values. Then, we tried modifying some parameters, getting the
following results:

           QUIET
   TXNBYTELIMIT  2048
    SKIPLANSERVERACP  YES

 with these values, the delayed time was of 4h 39 minutes.

                  QUIET
   TXNBYTELIMIT  25600
                  TXNGROUPMAX  256

 with these ones, the delayed time was of 4 h and 46 minutes

and finally, with the following parameters:

                 MEMORYEFFICIENTBACKUP YES
   TXNBYTELIMIT  25600
                  TCPBUFFERSIZE  32
                  TCPWINDOWSIZE 24
                  SKIPLANSERVERACP YES

we got 5 h 16 minutes.

  We also installed version 3 client to see whether we could improve it, but
it
continued the same.

  Now we are trying to make a test with another configuration, such as a
Token
Ring network attached directly to VSE, but we haven't been able to do it
yet.
However, which is the performance I should get with this kind of
configuration?.

  In the meantime, the customer has bought Internal Drives for 4mm tapes,
and
we also have tested the backup. In this case, you already know the results,
1.1
GB in 1 h and 38 minutes. This is OK compared to LANRES results, but it's
still
slow. This last result was tested with the values that best performance gave
with LANRES.

  Tomorrow I'll try again changing some parameters, and I'll let you know
the
result of the command q db f=detail.

  Lots of thank you in advance,

Mariajo.

Marma Josi Rodrmguez Canales
Dpto. Gestisn de Sistemas y Redes - IBM
e-mail: mar_jose AT es.ibm DOT com
Tf:  91 - 397 70 19
=
<Prev in Thread] Current Thread [Next in Thread>
  • [no subject], Maria Jose Rodriguez Canales <=