home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!uvaarpa!darwin.sura.net!paladin.american.edu!auvm!USCMVSA.BITNET!LDW
- Message-ID: <IBMTCP-L%93012805091774@PUCC.PRINCETON.EDU>
- Newsgroups: bit.listserv.ibmtcp-l
- Date: Thu, 28 Jan 1993 02:10:00 PST
- Sender: IBM TCP/IP List <IBMTCP-L@PUCC.BITNET>
- From: Leonard D Woren <LDW@USCMVSA.BITNET>
- Subject: Re: IBM TCP/IP inadequacies
- Lines: 64
-
- (You can never re-can a can of worms using the same size can...)
-
- On Wed, 27 Jan 1993 10:30:32 EST,
- Gary Buhrmaster <TJF@CORNELLC.CIT.CORNELL.EDU> said:
- > On Wed, 27 Jan 1993 01:54:00 PST, Leonard D Woren <LDW@USCMVSA.BITNET>
- > said:
- > >On Tue, 26 Jan 1993 13:09:34 CST,
- > > Rick Troth <TROTH@RICEVM1.RICE.EDU> re-opened a can of worms:
- > >> whatever is native to MVS inter-process comm when running on MVS.
- > >
- > >Well, there really isn't one, so every product has to invent their own.
- > >This is one of the most glaring deficiencies in MVS.
- > >
-
- > Well, actually, IBM seems to be promoting APPC for communication, which
-
- Bletch. Another complex thing with an incomprehensible programmer
- interface. APPC/MVS needs to speak TCP/IP; the APPC/MVS product
- people don't seem to understand this. (Also, there should be another
- TIOC to support TSO logon through TCP/IP without going through VTAM.
- And JES2(RSCS) should support JES2-JES2(RSCS) connections over TCP.
- How much of this is likely to happen? Why do I get the idea that most
- of IBM regards TCP/IP as a wart on the side of MVS?)
-
- > would allow you to run the server elsewhere in the gaggle of (smaller,
- > cost effective) MVS/ESA machines in your sysplex. However, when you
-
- I'm sure there's an oxymoron in there somewhere.
-
- > can be assured the server and client are within the same physical
- > machine, the architected solution would seem to be AR mode where the
- > server could directly access the private area(s) of the client, or a
- > special TCP/IP dataspace used to exchange data between the client and
- > the server using normal instructions without all the extra DAS
- > overhead of moving data between the secondary and primary address
- > spaces. The performance could be outstanding. As I recall that was
-
- Yeah, right, all well and good, except for one thing. AR mode
- requires that you have an active addressing bind to the other
- addressing space, or that you have it's local lock, or that it be
- non-swappable. These are generally Bad Things (tm).
-
- > one of the big performance gains in the original UNIX TCP/IP
- > implementations (passing pointers to the buffers rather than moving the
- > buffer data). Unfortunately, these features only work at current
- > MVS/ESA levels, and IBM seems committed to continuing TCP/IP support
- > for MVS/XA, which means they are limited to using facilties that it
- > supports.
-
- "I pray daily that programmers will be freed from the curses of
- compatibility." -- Edgar Dykstra [I hope I remembered this correctly.]
-
- It sure would be nice if MVS TCP/IP exploited some MVS/ESA features as
- suggested above. To do this IBM would either have to have dual path
- code, or drop support for pre-ESA MVS. I suspect there are enough XA
- shops who would scream too loud for the latter to be a possibility.
- For the former, in some cases there might be features which wouldn't
- work pre-ESA. But I think it's appropriate to try to provide the best
- possible performance when running on a current MVS release.
- Comments, IBM?
-
- /Leonard
-
- P.S. Hmmm... how about connecting two MVS TCPIPs via XCF?
-