PI52549: WMQ Z/OS V710:INCREASED STORAGE DUE TO READAHEAD PROCESSING

Obtain the fix for this APAR.

APAR status

• Closed as program error.

Error description

• Contention between Deferred write processing and Readahead
  processing is causing both delays to deferred write processing,
  and an increase in  requests for readahead processing. This in
  turn has resulted in an  increase in storage required by
  readahead, and in this instance, caused storage exhaustion .
  
  In MSTR joblog,the following messages are displayed:
  
  01.28.38 S0129303  CSQY222E +CSQ1 CSQSCTL Queue manager is
  critically  short of local storage - take action
  
  ...
  
  01.33.44
  IEA794I SVC DUMP HAS CAPTURED:
  
  DUMPID=056 REQUESTED BY JOB (CSQ1MSTR)
  
  DUMP TITLE=CSQ1,ABN=5C6-00E20003,U=SYSOPR
  ,C=R3600.710.BUFF-CSQP1RAH,M=CSQGFRCV,LOC=CSQSLD1 .CSQSVBK
  +00000C44
  
  Additional Symptom(s) Search Keyword(s):
  performance cpu I/O MMRE
  

Local fix

• n/a
  

Problem summary

• ****************************************************************
  * USERS AFFECTED: All users of WebSphere MQ for z/OS Version 7 *
  *                 Release 1 Modification 0.                    *
  ****************************************************************
  * PROBLEM DESCRIPTION: Various problems encountered when queue *
  *                      readahead is enabled (for example, when *
  *                      using QREP).                            *
  *                                                              *
  *                      Problems can include:                   *
  *                      - High CPU utilisation by the queue     *
  *                        manager address space, in some cases  *
  *                        leading to abend S071-20 and abnormal *
  *                        queue manager termination             *
  *                      - High number of IO requests to the     *
  *                        pageset datasets                      *
  *                      - Increase in queue manager storage     *
  *                        associated with MMRE control blocks,  *
  *                        potentially leading to a Short On     *
  *                        Storage condition and abnormal        *
  *                        queue manager termination             *
  ****************************************************************
  * RECOMMENDATION:                                              *
  ****************************************************************
  When using readahead, getting tasks queue requests to the
  readahead task so that it can bring in the pages required by the
  getter ahead of them being requested, and consequently reduce
  pageset i/o delays in the getting task.
  However, if the readahead task is delayed it can fall behind the
  getting task. This leads to an increase in the number of
  requests queued to the readahead task, including multiple
  requests for the same pages.
  When processing the requests, the large number of duplicated
  page requests can cause CSQIRAHP to loop scanning a list of
  pages for processing, leading to high cpu on this SRB.
  
  When the readahead task finishes processing the requests, or
  finds a request requiring IO to satisfy, it will attempt to
  load those pages, leading to further cpu cost if the pages
  are already in buffers, or IO requests if they are not. However
  the majority of these requests are for pages no longer needed
  by the getting tasks.
  
  If requests are being queued to the readahead task faster than
  it can process them, there can be a considerable delay in MMRE
  blocks associated with completed IOs being freed. This leads
  to an increase in the storage required for MMREs, due to the
  MMREs for completed IOs not being available for reuse by new
  IO requests.
  

Problem conclusion

• CSQIRAHP is changed to detect requests that are no longer
  current, and avoid doing unnecessary processing and IO for such
  requests. In addition CSQIMGE9 is changed to only queue a new
  request for a given page to the readahead task if that page is
  not already in the process of being loaded by the readahead
  task.
  100Y
  CSQIMGE9
  CSQIRAHP
  CSQPMTRM
  CSQP1GET
  CSQP1RAH
  

Temporary fix

• *********
  * HIPER *
  *********
  

Comments

APAR Information

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

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

  UI33764

Modules/Macros

• CSQIMGE9 CSQIRAHP CSQPMTRM CSQP1GET CSQP1RAH
  

Fix information

• Fixed component name

  WMQ Z/OS V7

• Fixed component ID

  5655R3600

Applicable component levels

• R100 PSY UI33764

     UP16/01/08 P F601 ¢

Fix is available

• Select the PTF appropriate for your component level. You will be required to sign in. Distribution on physical media is not available in all countries.