HP OpenVMS Systems

ask the wizard
Content starts here

Mailbox, Network Tasks Within Cluster?

» close window

The Question is:

I am creating a permanent mailbox on one host in a cluster. I need to obtain
 the particulars of the device from other hosts in the same cluster. Sure, this
 information could be written out to a file by the creating host, but I would
 much prefer getting th
e information interactively from the system tables. How can I do this from the
 DCL context?

The Answer is :

  Mailboxes can be an inappropriate solution for distributed cluster
  communucations, please use a network transport such as ICC (part of
  OpenVMS), DECnet, TCP/IP, or UDP/IP.
  While remote access to a mailbox is possible using RMS, DECnet, and FAL,
  there are definite limitations with this approach and there are some
  interesting requirements around message I/O -- due to file-oriented
  assumptions within RMS -- and for this and for other, the OpenVMS
  Wizard simply cannot recommend the use of mailboxes for anything but
  forever-local and datagram-oriented communications.
  The approach of writing information such as a mailbox unit number or
  other information into logical names or other static storage tends to
  lead to stale information -- if you must maintain one and only one
  copy of data, then a lock value block can be an appropriate solution.
  (Approaches such as DNS/Bind or other name services, or the use of
  network transports and servers based on DECnet task-to-task or TCP/IP
  avoid these limitations.)  If you really want to use a logical name
  (and you don't mind the potential for stale information), then the
  use of the cluster-wide logical name table is probably the easiest
  approach for this application.
  The OpenVMS Wizard also tends to prefer to avoid permanent mailboxes,
  since every mailbox should always have at least one associated server
  process reading from the mailbox -- thus a temporary mailbox works as
  well as a permanent mailbox, and a temporary mailbox cleans up itself.
  Also in general, there should be exactly one reader of any particular
  mailbox -- and zero or more writers -- as protocol bugs are common with
  most bidirectional mailbox communications schemes.  (The turn-around
  and the correct routing of the communications traffic is a particularly
  interesting area when using bi-directional mailbox traffic and multiple
  Examples of servers are in TCP/IP Services area TCPIP$EXAMPLES: and
  in SYS$EXAMPLES:DB*.*, and in available product documentation.
  If you insist on using mailboxes and logical names for this, you can
  obtain the logical name translation on a remote node using DCL and
  DECnet task-to-task communications.
  Please see topic (159) for details of DCL DECnet task-to-task.
  Please see topics such as (1910) and (6357) for some more general
  network communications (application) discussions.

answer written or last revised on ( 19-DEC-2001 )

» close window