ITEM: BT8052L

repeated VSD_CALL_ER messages on Oracle Parallel Server nodes



Question:

ENV:  AIX Level:      4.1.4, PSSP 2.1.0.17

DESC:

We get massive numbers of VSD_Call_Err errors on many of our SP nodes
which are running Oracle Parallel Server.  OPS registers NO ORA errors.
We are taking no hits on the switch or switch communications.  Running
a de-config & reconfig/restart of the VSD comes up 'clean' (no errors).

The Following is a clip from the error log on one node that takes the
VSD error hits:
---------------------------------------------------------------------------
LABEL:          VSD_CALL_ER
IDENTIFIER:     EA97C090

Date/Time:       Wed Feb  5 09:24:10
Sequence Number: 5096
Machine Id:      000211635900
Node Id:         sp2-node28
Class:           S
Type:            PERM
Resource Name:   vsdd

Description
SERVICE CALL FAILED

Probable Causes
SUBSYSTEM

User Causes
SERVICE CALL CANNOT COMPLETE FOR

        Recommended Actions
        VERIFY SYSTEM CONFIGURATION IS VALID
Failure Causes
UNEXPECTED RETURN CODE...SERVICE CALL

        Recommended Actions
        PERFORM PROBLEM DETERMINATION PROCEDURES

Detail Data
DETECTING MODULE
LPP=PSSP,Fn=todo.c,SID=1.31,L\#=614
RETURN CODE
GetCommAdapterBuffer: IFIOCTL_MGET(): data NULL
--------------------------------------------------

The last line, IFIOCTL_MGET(): data NULL  is indicitive
of a communications error, but we can't see that the switch is having
problems.

We have PTF 18 of PSSP 2.1 loaded and the switch is a tb3 switch.

Response:

The rpoolsize and spoolsize are set too low.  The buffer on the switch
is filling up and causing the VSD calls to fail.  By default the
rpoolsize and spoolsize are set to .5M.  This is generally too small
for most VSD systems.  /usr/lpp/ssp/css/vdidl3 -i shows several
failures in the send/recv pools.  This indicates the buffer sizes
are too smal.  The minimum size needs to be the size of
tcp_sendspace or tcp_recevspace, which ever of the two is larger.
A good size to start out with is 5-6MB for rpoolsize and spoolsize
and then bumping up from there if need be.
.To change the values run:
        chgss -l css0 -a spoolsize\ -a rpoolsize=\
and reboot the nodes.
.The command /usr/lpp/ssp/css/vdidl3 is an undocumented command
used by development.  However, end-users can use this command to
see how many "failures" they are receiving.  The command looks at
the switch device driver and any failures indicates the buffer filled
up your rpoolsize/spoolsize may be too small.


Support Line: repeated VSD_CALL_ER messages on Oracle Parallel Server nodes ITEM: BT8052L
Dated: February 1997 Category: N/A
This HTML file was generated 99/06/24~13:30:19
Comments or suggestions? Contact us