Performance Difference PDL TCP/IP Versus Native TCP/IP

This topic has 3 replies, 2 voices, and was last updated 15 years, 8 months ago by Jim Kosloskey.

Creator

Topic
April 13, 2009 at 5:42 pm #50792
Bob Richardson
Participant
Greetings,

We are an AIX 5.3 TL8 SP3 shop running CIS5.6 Revision 2, multi-site configuration. To communicate among these sites, we use localhost bridges.

During our testing for CIS5.6 Rev2 (we migrated from CIS5.3 Rev2), we observed that using the auto-reply/resend feature for our localhost bridge connections (we had to drop the old recovery_33 procs), we lost messages when the threads and/or remote connections were shutdown or cycled. We had configured these bridges as Native TCP/IP for efficiency purposes. To get the messages in State 16 to persist over shutdown sequences, we had to re-configure them as PDL style.

We are in the midst of a project to identify tuning opportunities on our engines and we *think* that the Native TCP/IP would be more efficient in resources used (higher performer) than PDL style. To implement the Native option would require that we crank out a couple of reply/resend procedures however as auto-reply/resend would not work (based on our experience). Upgrading to CIS5.7 where this is fixed is not an option until next year for us (per management).

Question: based on your experiences when configuring localhost bridges among multiple sites, do you use Native or PDL TCP/IP?

Your feedback would be most welcome.

Thanks in advance.
Creator

Topic

Viewing 2 reply threads

Author

Replies
- April 13, 2009 at 7:20 pm #67540
  Jim Kosloskey
  Participant
  Bob,
  
  There is a performance improvement using length encoded TCP/IP versus PDL. I have never measured it but I think the code to do that could be fairly straight forward.
  
  In a previous life, we used length-encoded TCP/IP between sites and processes.
  
  Here we use PDL. The reason is there is a resistance to seeing the log posts when the TCP/IP Connection is doing its reconnects when the server side is stopped.
  
  Personally I don’t think that is a big issue and would prefer TCP/IP. Still need the acknowledgments as you have discovered.
  
  It is interesting there should be a loss of messages using length encoded TCP/IP with the new default acknowledegment handling.
  
  On the surface, this sounds like a bug to me.
  
  Has this been brought to support’s attention?
  
  If so are they considering it a bug or is the automatic recovery only intended for HL/7?
  
  email: jim.kosloskey@jim-kosloskey.com 29+ years Cloverleaf, 59 years IT - old fart.
- April 13, 2009 at 8:33 pm #67541
  Bob Richardson
  Participant
  Greetings again,
  
  The problem with Native TCP/IP not saving the State 16 across shutdown events was reported during our evaluation of CIS5.7. My *assumption* at that time was that this “problem” was preserved across versions going from CIS5.6 to CIS5.7. It was fixed in 5.7 (we received a Beta 2 version) per my tests. To my knowledge, it is not an official bug for CIS5.6; most likely a shop would need to upgrade to 5.7 for a fix, that is, if they notice a loss of messages when auto reply/resend is configured and they are using Native TCP/IP. My *guess* is that the “auto” feature was included in the PDL driver logic only as most shops use HL7 and PDL. The alternative would be to write some reply resend logic like in the old recovery_33 days and reference the new message key OBMSGID instead of the State 14 saved message handle.
  
  Let me know if you need more information.
  
  Thanks for the response.
- April 13, 2009 at 9:07 pm #67542
  Jim Kosloskey
  Participant
  Bob,
  
  Thanks – that is good information.
  
  email: jim.kosloskey@jim-kosloskey.com 29+ years Cloverleaf, 59 years IT - old fart.
Author

Replies

Viewing 2 reply threads

The forum ‘Cloverleaf’ is closed to new topics and replies.