OW52291: IMPLEMENT PACKING FOR ENTERPRISE EXTENDER

Obtain the fix for this APAR.

APAR status

• Closed as unreproducible in next release.

Error description

• Performance degraded when running EEover QDIO or iQDIO,
  or native TCP/IP traffic (not EE) over QDIO.
  Additional symptoms:
  Recursive abend module IVTSMCFB called by ISTLLCFB called
  by ISTLLCWI. The RPH is not removed from the PAB.
  

Local fix

Problem summary

• ****************************************************************
  * USERS AFFECTED: Two types of users:                          *
  *                   1. All OSA Express QDIO or HiperSockets    *
  *                      (iQDIO) users                           *
  *                   2. Users of Enterprise Extender (EE) over  *
  *                      OSA Express QDIO or over HiperSockets   *
  *                      (iQDIO)                                 *
  ****************************************************************
  * PROBLEM DESCRIPTION: Performance degraded when running EE    *
  *                      over QDIO or iQDIO, or native TCP/IP    *
  *                      traffic (not EE) over QDIO.             *
  ****************************************************************
  * RECOMMENDATION:                                              *
  ****************************************************************
  For some configurations and system workloads, system performance
  including CPU utilization, throughput, latency, and transactions
  per second can be less than optimum when running EE over QDIO
  or iQDIO.
  
  Currently, EE packets are not packed into contiguous storage.
  The SNA upper layers pass user data down to the HPR layer. The
  HPR processing will then preappend the HPR headers. The EE
  processing will then pre-append the UDP and any necessary media
  headers.
  
  This results in 3 small elements (headers or buffers) followed
  by the user data (SNA RU). Often, the user data can also be in
  multiple small pieces.
  
  Since this data is not packed into contiguous storage, the QDIO
  DLC must then process each individual buffer (XBFLENT) by
  mapping each into a 4k buffer entry (SBALE).
  
  This can become wasteful, resulting in excessive I/O operations
  (i.e. when each single packet of 1492 bytes could have been
  transmitted using a single 4k SBALE vs. using 4 SBALEs).
  

Problem conclusion

Temporary fix

Comments

• A Communications Server enhancement has been provided by this
  APAR to now perform packing for EE over QDIO or iQDIO.
  Each NLP (Network Layer Packet) will be moved into a 1500 byte
  contiguous storage area prior to presenting the packet to the
  OSA or HiperSockets (iQDIO) adapter.
  
  This will result in a reduced number of I/O operations (SIGAs),
  and more efficient usage of each operation or write buffer
  (SBAL). OSA configurations using a 1492 Ethernet frame size
  will result in a single packet per QDIO SBALE (vs. 4 or 5).
  Data (MTU) that exceeds 1500 bytes (e.g. 9k Jumbo frames) will
  not be packed beyond 1520 bytes (i.e. we will pack up to 1520,
  and transmit the remainder as is (without moving)).
  
  There are no external definition or configuration changes
  required to enable this improvement.
  
  A new VTAM GETBLK (TIPACPK) pool was created. VTAM GETBLK
  storage can be monitored using D Net, VTAMSTOR.
  
  To provide this improvement, the following changes are made:
  
    1) A new VTAM GETBLK pool called TIPACPK is created. These
       buffers are similar to TIPACs and TIPACX, and come from
       fixed ECSA storage. They are "packing TIPACs" (or TIPACs
       with a "packing area extension").
       Parts ISTORCEK, GETPINFO, RAFTSMS, ISTDT033, and ISTTT007
       were changed to create the new pool.
  
    2) Parts ISTRPCRS, ISTTSCMO, ISTMAPPL, ISTTSCUC were changed
       to use the new packing TIPACs from new pool TIPACPK.
  
    3) ISTLLCWI has been changed to provide EE packing support.
  
       OSA Development has provided a Gigabit Ethernet performance
       improvement that will reduce the number of dropped outbound
       packets. Ths OSA fix levels which contain the fix are:
  
                        G5/G6: 4.24 (or above)
            z/Series 2064 GA1: 1.49 (or above)
                          GA2: 2.26 (or above)
  
       Although the OSA change is not required, it is recommended
       that EE users upgrade their OSA Express microcode to
       include that change prior to exploiting this Communications
       Server PTF.
  
       Note that the recommendation only applies to EE over OSA
       Express using Gigabit Ethernet (i.e. the OSA solution does
       not apply to Fast Ethernet).
  
  A second improvement has been provided that applies to both OSA
  Express QDIO and HiperSockets users.
  
  In order to handle the increased bandwidth of both OSA Express
  QDIO and HiperSockets (iQDIO), the DLC layer (or device driver)
  will now increase the amount of storage used for read (inbound)
  processing.
  
  Analysis of performance data of high volume workloads at the
  higher data rates (gigabit) has demonstrated that the adapter
  was overrunning the number of available read buffers.
  This results in dropped packets and poor overall performance.
  The storage increase is as follows:
  
         1. HiperSockets:
  
              CHPID MFS         old        new
                64k            4 meg      8 meg
                40k            4 meg      5 meg
                24k            3 meg      3 meg  (no change)
                16k            2 meg      2 meg  (no change)
  
         2. OSA Express :
                                old        new
             zSeries (64bit)  .5 meg      4 meg
             non 64bit        .5 meg      1 meg
  
      This storage is allocated from the 4k CSM data space pool,
      and is on a per active QDIO (or iQDIO) data device basis.
      This storage is page fixed.
  
  It is recommended that users review their CSM usage and
  IVTPRM00 parmlib specifications for fixed CSM storage. Refer to
  the MVS Tuning and Initialization Reference SA22-7592-01 Chapter
  55 IVTPRM00 for details.  Information APAR II12657 provides
  additional CSM general information.
  
  Because of this storage increase, this APAR also provides two
  new VTAM start options which allow the user to configure the
  amount of storage used. For most users the IBM defaults will be
  sufficient. The new options are:
  
           QDIOSTG  (QDIO Storage)
           IQDIOSTG (iQDIO Storage)
  
          _QDIOSTG=_MAX________
     >>__|_____________________|________><
         |_QDIOSTG=_MAX________|
                  |_AVG________|
                  |_MIN________|
                  |_NNN________|
  
          _IQDIOSTG=_MAX_______
     >>__|_____________________|________><
         |_IQDIOSTG=_MAX_______|
                  |_AVG________|
                  |_MIN________|
                  |_NNN________|
  
  For the usage details of the two new options, reference
  information APAR II12488 (SNA Resource Definition Reference).
  
  This APAR is a logical route of the following development items:
  
         D131137 (EE packing implementation)
         D131538, D131867 (QDIOSTG option implementation)
  
  D131456 is also partially included.
  
  ISTLLCD3 has been included for maintenance purposes.
  

APAR Information

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

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

  UA00131 UA00132

Modules/Macros

• GETPINFO ISTCICPR ISTCICS2 ISTDT033 ISTINCEI
  ISTLLCD3 ISTLLCFB ISTLLCWI ISTMAPPL ISTNOCPR ISTORCEK ISTORCGB
  ISTRPCRS ISTTSCMO ISTTSCUC ISTTSEXT ISTTT007 ISTXBFL  IUTLLCCC
  IUTLLCRU IUTXBFLC RAFTSMS
  

Fix information

• Fixed component name

  VTAM V4 MVS/ESA

• Fixed component ID

  569511701

Applicable component levels

• R120 PSY UA00131

     UP03/03/06 P F303

• R140 PSY UA00132

     UP03/03/06 P F303

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.