OA39261: HANG ASYNCIO EXIT SRB SCHEDULING OSI_WAIT BLOCKED SYNCHRONOUS

Obtain the fix for this APAR.

APAR status

• Closed as program error.

Error description

• An application issues an asyncio read operation. When data is
  received TCPIP issues an osi_schedule and the specified
  application exit is invoked on an SRB in the z/OS UNIX SRB
  pool to complete the read processing.
  As a part of its read completion processing,
  the exit issues a (synchronous) write syscall.
  The write blocks in TCPIP due to outbound buffer shortage;
  the SRB is placed in an osi_wait.
  
  Subsequently a previously issued asyncio write completes in
  TCPIP, which issues an osi_schedule to schedule the application
  I/O completion exit.  The osi_schedule processing does not
  recognize that the prior read completion exit is in an
  osi_wait and schedules the asyncio completion for the write
  to run on the same SRB.  This results in a delay until the
  osi_schedule processing subsequently recognizes that pending
  work is not being processed and schedules new SRBs to process
  the asyncio completion events.
  
  ** Note that R770 PTF UA65441 and R780 PTF UA65442 for this APAR
     have been marked as PE.  The fixing APAR is OA44221.
  

Local fix

Problem summary

• ****************************************************************
  * USERS AFFECTED: Users of z/OS Unix System Services           *
  *                 for HBB7760, HBB7770, and HBB7780            *
  *                 using WebSphere Application Server z/OS      *
  *                 at level 6.1.0 or higher.                    *
  ****************************************************************
  * PROBLEM DESCRIPTION:   A slowdown is noticed in the          *
  *                      performance of a network server         *
  *                      that uses z/OS Unix Asynchronous I/O.   *
  *                        WebSphere Application Server is       *
  *                      susceptible to this problem if there    *
  *                      is congestion on the network that it    *
  *                      is using.                               *
  ****************************************************************
  * RECOMMENDATION:                                              *
  ****************************************************************
    When an Async I/O Completion Exit causes the z/OS Unix SRB
  that it is running on to wait, the SRB Scheduling algorithm
  in BPXPRSSR is not aware that this SRB is hung up and so not
  really available for new work.
    This causes new osi_sched's to be queued up until there are
  five or more requests queued or after several minutes when a
  timer deamon in the Kernel notices that the queue is not being
  processed.
    This delays the running of Asyncio Part 2 and the completions
  of I/Os that have been started, which is observed as a slowdown
  in the server application.
  

Problem conclusion

• When an Asyncio Exit makes a syscall that has to wait, the
  z/OS Unix SRB that it is running on will be removed from the
  pool of SRBs available for future I/O Complete Notifications.
  
  This will cause a new SRB to be scheduled to handle the next
  I/O Completion that occurs and there will be no delays.
  

Temporary fix

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

Comments

• ×**** PE14/01/14 FIX IN ERROR. SEE APAR OA44221  FOR DESCRIPTION
  

APAR Information

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

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

  UA65440 UA65441 UA65442

Modules/Macros

• BPXFOWT  BPXPRSSR
  

Fix information

• Fixed component name

  OPENMVS SYS SRV

• Fixed component ID

  5695SCPX1

Applicable component levels

• R760 PSY UA65440

     UP12/06/06 P F206 Ž

• R770 PSY UA65441

     UP12/06/06 P F206 Ž

• R780 PSY UA65442

     UP12/06/06 P F206 Ž

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.