PM82672: IPIC PERFORMANCE IMPROVEMENT WITH LOCK WAIT REDUCTION

APAR status

• Closed as program error.

Error description

• You are running a stress workload to cross instance applications
  with RTIMOUT and MONITOR timeouts set to purge transactions that
  take excessive time to complete.  When one region is slowed, the
  processing is affected on the other regions, as multiple tasks
  are delayed and eventually purged.
  .
  Investigation of dumps reveal that much time is spent by tasks
  waiting for the exclusive lock for an INQUIRE call for DFHSOTB.
  The information sought from the INQUIRE is the port number of
  the TCPIPSERVICE.  This information is static, and could be
  obtained and saved during CAPEX processing, avoiding the lock
  entirely at this juncture.
  KIXREVEAA
  

Local fix

Problem summary

• ****************************************************************
  * USERS AFFECTED: All CICS users.                              *
  ****************************************************************
  * PROBLEM DESCRIPTION: CICS region slowdown after a large      *
  *                      number of IPIC DPL requests are         *
  *                      received.                               *
  ****************************************************************
  * RECOMMENDATION:                                              *
  ****************************************************************
  A large number of IPIC DPL requests are received in CICS.
  Mirror tasks are attached in the remote region to handle them.
  When each mirror task starts, DFHISXM runs for init and bind.
  During bind, a monitoring performance record is captured.
  This record requires the TCPIPSERVICE name and port number.
  The port number is not held by IS domain so a DFHSOTB
  INQUIRE_TCPIPSERVICE call is made.
  DFHSOTB acquires the TCPIPSERVICE lock exclusively, and then
  gathers the information required to satisfy the inquire (by
  calling DFHSOTD).  This does not check which fields were
  requested so does a lot of processing, including switching
  to the SO TCB as part of inquire_connection.
  While on the SO TCB the next mirror task can start on QR.
  It reaches DFHSOTB and then suspends waiting for the lock.
  In a busy system many mirror tasks can end up waiting.
  The first mirror task returns to QR and releases the lock.
  The other mirror tasks all resume and retry the lock acquire.
  This causes excessive processing as tasks suspend and resume.
  Due to the nondeterministic nature of lock acquire, some tasks
  may time out after multiple attempts to acquire the lock.
  

Problem conclusion

• Copybook DFHISDCC is changed to add field iscb_tcpips_port to
  the ISCB (IPCONN) control block.
  Module DFHISXM is changed to store the TCPIPSERVICE port number
  in iscb_tcpips_port for the applicable ISCB.
  Program DFHISCOP is changed to clear iscb_tcpips_port in the
  ISCB when the associated IPCONN is released.
  

Temporary fix

• FIX AVAILABLE BY PTF ONLY
  

Comments

APAR Information

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

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

  PM88910

Modules/Macros

•    DFHCIS4  DFHISAL  DFHISCO  DFHISCOP DFHISCU
  DFHISDM  DFHISDUF DFHISEM  DFHISEMP DFHISFS  DFHISIC  DFHISIF
  DFHISIS  DFHISJU  DFHISLQ  DFHISLQP DFHISRE  DFHISRR  DFHISRRP
  DFHISRS  DFHISRSP DFHISSR  DFHISST  DFHISTRI DFHISUE  DFHISXF
  DFHISXM  DFHISXS  DFHISZA  DFHMIRS
  

Fix information

• Fixed component name

  CICS TS Z/OS V4

• Fixed component ID

  5655S9700

Applicable component levels

• R700 PSY

     UP