IC70828: SLOW TCP/IP CONNECTION FROM TPC TO EMC CIMOM

APAR status

• Closed as Permanent restriction.

Error description

• Problem was in negative impact of Nagle algorithm ( see
  http://en.wikipedia.org/wiki/Nagle's_algorithm)
  on CIMOM traffic.
  Due to nature of CIMOM request/responce exchange (in customer
  case there were about 50k reqs per EMC device) there were a lot
  very small TCP packets between TPC server and EMC SE hosts.
  In case of AIX Nagle algorithm could add up to 500ms delay to
  every TCP packet send to CIMOM server.
  
  By default on AIX Nagle algorith turned on for packets less than
  64KB (my assumption is that all CIMOM request is affected by
  Nagle algorith).
  

Local fix

• Problem was fixed chenging the following parameters:
  
  no -p -o tcp_nodelayack=1
  http://publib.boulder.ibm.com/infocenter/aix/v6r1/topic/com.ibm.
  aix.prftungd/doc/prftungd/tcp_nodelayack.htm
  
  tcp_nodelay=1 on interface which connected to CIMOM
  http://publib.boulder.ibm.com/infocenter/aix/v6r1/topic/com.ibm.
  aix.prftungd/doc/prftungd/tcp_nodelay_isno.htm
  

Problem summary

• ****************************************************************
  * USERS AFFECTED: TPC running on AIX Operating systems and     *
  *                 probing CIM devices.                         *
  ****************************************************************
  * PROBLEM DESCRIPTION: See ERROR DESCRIPTION                   *
  ****************************************************************
  * RECOMMENDATION:                                              *
  ****************************************************************
  AIX has the TCP/IP option "TCP_NODELAY" disabled per default
  for optimization purposes.
  For TCP connections like CIM calls TPC is using for device
  probes, this setting increases the connection time and slows
  down the probe performance.
  A solution to change the network setting by TPC itself will be
  tracked in a separate defect,
  because underlying components (Java for AIX) has to be fixed.
  

Problem conclusion

• Workaround:
  On the TPC server the option TCP_NODELAY can be enabled manually
   on the AIX OS level using command:
  chdev  -l 'en0' -a state='up' -a tcp_nodelay='1'
  where "en0" is the name of the Network interface used to connect
   to the CIM device.
  With command
  lsattr -E -l en0 |grep tcp_nodelay
  the value for TCP_NODELAY can be verified and should be reported
   with "1" by AIX.
  
  Java and AIX fixes may be available at a later date.  This is cu
  rrently a limitation that Tivoli Storage Productivity Center has
  .
  

Temporary fix

Comments

APAR Information

• APAR is sysrouted FROM one or more of the following:

• APAR is sysrouted TO one or more of the following:

Fix information

Applicable component levels