HP OpenVMS Systems

ask the wizard
Content starts here

Networking, Application Performance?

» close window

The Question is:

 Is there a lot of system overhead when using
 TCP/IP socket IPC between 2 processes on the
 same local machine. What I mean is , is it justifiable (and efficient) to use
 sockets in place of mailboxs or other local IPC methods
on VMS.

The Answer is :

  There is no certain answer without rather more detail of course, but
  the OpenVMS Wizard would not expect significant intra-node performance
  differences among the various inter-process communications techniques,
  once the initial connections (for the connection-oriented protocols)
  are established.  (Connectionless mechanisms can have advantages over
  connection-oriented protocols for certain communications traffic.)
  Within an OpenVMS Cluster, the OpenVMS Wizard would typically first look
  at using ICC (IntraCluster Communications) and IP.
  One obvious reason for choosing a network protocol (or even middleware
  package) would involve using source code that would be independent of
  the particular communications patterns, allowing for possible future
  changes to the application -- permitting future expansion to multiple
  cluster members or multiple networked systems.
  A secondary reason might involve simple familiarity and overall
  expedience: programmer familiarity with networking techniques, and the
  relative ease of avoiding various software inefficiencies through the
  addition of hardware.  In other words, using techniques and tools
  familiar to the programmers and hardware upgrades may well be cheaper
  than using more efficient coding techniques and tools.  And it may be
  easier to evaluate the entire application system -- either in prototype
  form, or even once the baseline system is running -- for any required
  performance tuning or for hardware upgrades, if any are even required.

answer written or last revised on ( 21-SEP-2000 )

» close window