FCM AUTOMATIC setting isn’t fast enough?

October 8, 2013

ENVIRONMENT:

DB2 v10.1 FP 2
OS: RHEL 6.4

PROBLEM:

On one of our DPF environment, we noticed a resource intensive job failed exhausting both the FCM Buffers & channels. Would see the following entries in the db2diag.log and also errors about “No FCM Buffer Available”

2013-10-03-05.49.47.787008-300 I518288292E547 LEVEL: Error
PID : 8369 TID : 46968990590720 PROC : db2sysc 21
INSTANCE: db2admn NODE : 021 DB : DBNAME
APPHDL : 901-58904 APPID: 10.96.37.168.38337.131003104954
AUTHID : USER HOSTNAME: hostname
EDUID : 173784 EDUNAME: db2agntp 21
FUNCTION: DB2 UDB, fast comm manager, sqkfChannelManager::AllocChannel, probe:5
RETCODE : ZRC=0x85590029=-2057764823=SQLKF_NO_CHANNEL
“No FCM Channels available”

This happened even though the FCM parameters are set to AUTOMATIC.

db2 get dbm cfg | grep -i fcm
No. of int. communication buffers(4KB)(FCM_NUM_BUFFERS) = AUTOMATIC(8192)
No. of int. communication channels   (FCM_NUM_CHANNELS) = AUTOMATIC(16384)

Did notice messages about DB2 increasing both the values, so the automatic did seem to be working. And didn’t have any limits on INSTANCE_MEMORY and had over 70% of Physical memory available on the server as well.

RESOLUTION:

Increasing the initial value to a higher number along with AUTOMATIC helped get the job completed ok.

UPDATE DBM CFG USING FCM_NUM_BUFFERS 32768 AUTOMATIC
UPDATE DBM CFG USING FCM_NUM_CHANNELS 32768 AUTOMATIC

ROOTCAUSE:

When FCM parameters are set to AUTOMATIC on Linux, it can only be increased 25% above the initial value. Its not clear from the documentation, whether the 25% limit is for an initial increase or only 25% is allowed above the initial value.

tags: , , , ,
posted in DB2 by Visu

Follow comments via the RSS Feed | Leave a comment | Trackback URL

Leave Your Comment

 
Powered by Wordpress and MySQL. Theme by Shlomi Noach, openark.org