OA20553: SLOW NETWORK PERFORMANCE WHEN RUNNING Z/OS AS A GUEST OF Z/VM 5.2 or higher (z/VM 5.4), WITH A DEDICATED OSA ADAPTER.

Obtain the fix for this APAR.

APAR status

• Closed as unreproducible in next release.

Error description

• Slow network performance when running z/OS 1.8 as a guest of
  of z/VM 5.2 or higher (i.e. z/VM 5.4) with a dedicated OSA
  adapter.
  
  
  
  
  Keywords: 5655HAL00
  

Local fix

Problem summary

• ****************************************************************
  * USERS AFFECTED: All running QDIO or iQDIO devices on a       *
  *                 second level z/OS guest machine on VM.       *
  ****************************************************************
  * PROBLEM DESCRIPTION: Getting very long ping response time    *
  *                      running OSA with a z/OS guest on VM.    *
  ****************************************************************
  * RECOMMENDATION:                                              *
  ****************************************************************
  The SLSB "Read Processing" state tells OSA not to interrupt
  the device driver because it would look ahead for more read
  completions and processing until none is found.
  
  However, the DLC does not tell VM (via SIGAs) when it enters
  that state, though VM finds out indirectly later when a
  SIGAr (an implied SIGAs) is issued.  This does cause the OSA
  to eventually enter the "Suppress Interrupt" state so
  interrupts may be disabled at that point.
  
  Moreover, the DLC fails to tell VM to synch up the shadow tables
  when "Read Processing" is exited. This keeps the OSA from giving
  the DLC an interrupt because it stills sees the "Read
  Processing" state in the shadow table (instead of the more
  current "Read Reset" state in the DLC table).  This in turn,
  keeps the DLC from actively looking for additional read
  completions while waiting for the interrupt.
  
  The device does not hang because part of the OSA algorithm is to
  interrupt anyway if the number of read buffers is low, but there
  is significant latency as a result.
  

Problem conclusion

Temporary fix

Comments

• ISTLLCIE will issue SIGAs to tell VM to update its shadow tables
  after it first enters the "Read Processing" state. It will not
  tell VM again if it propagates the state as that is unnecessary.
  
  IUTLLCDQ will issue SIGAs when it attempts to look ahead and
  when the "Read Processing" state is exited.
  
  ISTLLCM8 has been added for maintenance purposes.
  
  The above changes are also documented under IBM internal number
  D140676.  This APAR also fixes a problem in ISTLLCIE with a bad
  UCB address being passed to the z/OS PTrace routine (IBM
  internal number D140049).
  

APAR Information

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

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

  UA34126 UA34127 UA34128

Modules/Macros

• ISTLLCIE ISTLLCM8 IUTLLCDQ
  

Fix information

• Fixed component name

  VTAM V4 MVS/ESA

• Fixed component ID

  569511701

Applicable component levels

• R160 PSY UA34126

     UP07/05/24 P F705

• R170 PSY UA34127

     UP07/05/24 P F705

• R180 PSY UA34128

     UP07/05/24 P F705

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.