home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.unix.programmer
- Path: sparky!uunet!gatech!psuvax1!vogon1!jdm1
- From: jdm1@vogon1.pasen.gov (Jonathan D McCown)
- Subject: SysV Message Queues - Use in TP - (rambling comment/question)
- Message-ID: <1992Nov12.154253.4188@vogon1.pasen.gov>
- Summary: SysV Message Queues - Use in TP - (rambling comment/question)
- Keywords: Bogus SysV Message Queues Whatever
- Organization: Vogon Poets Society
- Distribution: na
- Date: Thu, 12 Nov 92 15:42:53 GMT
- Lines: 49
-
-
-
- I've been fiddling with a client/server configuration based around
- System V message queues. Somewhere in my reading I found that their
- development had come out of some Bell Labs (or whatever) Transaction
- Processing work, and I thought it would bear investigation. This post
- is to solicit comments, suggestions or increased news traffic.
-
- My prototype looks like this:
-
- --------- <------- SRVQUE <------ Client-1...n (write requests to server)
- MSERV |
- ________| -------> CLIQUE ------> Client-1...n (read requests from server)
-
- Messages to the server are coded by message-type (part of message queue
- structure) as to their transaction type. In this way should multiple
- specialized servers be reading the same queue, no conflict arises as
- they "read their own thing", this can have interesting effects if one
- spawns additional servers to handle traffic increases.
-
- Responses from the server are enqueued with a message-type == PID of
- the client, so clients may either block on read or poll for their
- responses.
-
- Since the processing is all "in core" the prototype is wonderfully fast.
- Initially ~230 "transactions" per second, using a non-optimized single
- threaded server. Naturally, adding IO heavy functions to the server slows
- things down. (I am linking in Informix ESQL funcs).
-
- I realize that this may not be bleeding edge for some, but it is an
- interesting study (and actually approaches a solution for some system
- bottlenecks in our environment). In my case, a lot of Informix system
- overhead can be overcome via one "informix q-server" servicing hundreds
- of _inquiry_ based applications. (informix has an "engine" proc for each
- client app in its standard implementation, many many many procs)
-
- Comments, suggestions welcome. Multi-threading would be great here...
- My environment is a Sequent (4 CPUs) under PTX 1.3 so some of the Sequent
- concurrent processing stuff may become involved (yeah..!).
-
- - J.D. McCown
- "It ain't CICS.. yet."
-
- Domain in transition, best mailpath may be "psuvax1!vogon1!jdm1"
-
- --
- J.D. McCown - Computer Services - Senate of Pennsylvania
- psuvax1!vogon1!jdm1 "You don't want to get locked into
- jdm1@vogon1.pasen.gov 'open systems'." -anonymous IBM Rep.
-