PQ99704: TCP SESSION HANGS - DATA SENT BEYOND WINDOW WHILE IN FRR

Obtain the fix for this APAR.

APAR status

• Closed as program error.

Error description

• TCP sessions that are established with a remote system that are
  using a high volume of outbound data traffic can occasionally
  hang with no progress.  If the local application uses a timeout
  function (such as FTP) or the remote application will timeout
  with a RESET, the session will terminate after that period of
  inactivity with a corresponding error indication (such as errno
  140=x'8C'=EPIPE).
  
  A packet trace (SYSTCPDA CTRACE) or sniffer trace of the session
  will show the following events:
   - Numerous ACKs from the remote side indicating missing or
     out-of-order packets.
   - Multiple of these in a row with the same ACK sequence number
     will trigger Fast Retransmit/Recovery (FRR) logic.
   - At the same time, the advertised window from the remote side
     shrinks, eventually becoming less than one MSS in size.
     Potentially, the remote side could even cut its window back
     to where the last acceptable sequence number is less than
     what was calculated from an earlier packet ("left shift" of
     the window).
   - One (or more) of the packet(s) transmitted due to FRR exceeds
     that window.  After this, no further data packets are
     transmitted (which includes window probes or other
     retransmits).
  

Local fix

• - Take actions to reduce the number of lost or out-of-order
    packets in the network between the z/OS system and the target
    server.
  - Increase the size of TCP buffers on the target server to avoid
    closing the advertised window while waiting for recovery of
    lost or out-of-order packets.
  - Use Policy Agent to regulate the rate of out bound packets to
    the target server.
  
  Other Symptoms:
  
    Print functions using TCP/IP are quite likely to be affected
    by this problem, since printer slow-downs (paper jam, out of
    paper, ...) make the server more likely to advertise a zero
    window.  These functions include LPR, Network Print Facility
    (NPF), PSF/MVS, VPS (*), and similar utilities.
  
  (*) (R) Levi, Ray and Shoup
  TCB_WRT_BLOCKED is set and the TCB_SND_WND is greater then
  zero. The TCB_IFR contains x'91' & x'92' entries.
  There is a write event element on the TCB event queue.
  Additional Symptoms: IFR91 IFR92
  

Problem summary

• ****************************************************************
  * USERS AFFECTED: All users of the IBM Communications Server   *
  *                 for z/OS Version 1 Release 2 IP              *
  ****************************************************************
  * PROBLEM DESCRIPTION: A TCP transmission that involves the    *
  *                      retransmission of data when the TCP     *
  *                      window is approaching zero may hang     *
  *                      and the transfer of data may not        *
  *                      complete.                               *
  ****************************************************************
  * RECOMMENDATION:                                              *
  ****************************************************************
  If a TCP retransmission of data occurs when the TCP window is
  less than a single packet in size, CS for z/OS TCPIP may
  retransmit a full packet that exceeds the window. If the
  receiver of such a packet does not ACK all of the data, CS for
  z/OS TCPIP may not send the next packet. As a result, TCP
  transmission will hang and will not successfully complete.
  The retransmission of data that would exceed the available
  window occurs because of code in EZBTCRD that sends a
  retransmission before updating the available window size.
  +-------------------------------------------------------------+
  + Please check our Communications Server for OS/390 homepages +
  + for common networking tips and fixes.  The URL for these    +
  + homepages can be found in Informational APAR II11334.       +
  +-------------------------------------------------------------+
  

Problem conclusion

• EZBTCRD has been amended to update the window size before
  attempting to retransmit data.
  
  * Cross Reference between External and Internal Names
  

Temporary fix

Comments

APAR Information

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

  PQ94985

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

  UQ97093

Modules/Macros

•    EZBTCRD
  

Fix information

• Fixed component name

  TCP/IP V3 MVS

• Fixed component ID

  5655HAL00

Applicable component levels

• R120 PSY UQ97093

     UP05/03/21 P F503

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.