home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!paladin.american.edu!auvm!USCMVSA.BITNET!LDW
- Message-ID: <IBMTCP-L%92122817204530@PUCC.PRINCETON.EDU>
- Newsgroups: bit.listserv.ibmtcp-l
- Date: Mon, 28 Dec 1992 14:20:00 PST
- Sender: IBM TCP/IP List <IBMTCP-L@PUCC.BITNET>
- From: Leonard D Woren <LDW@USCMVSA.BITNET>
- Subject: Re: IBM's MVS TCP/IP problems
- Lines: 71
-
- On Mon, 28 Dec 1992 11:09:31 EST,
- Ron Lane <RONLANE@YKTVMV.BITNET> said:
- > When I entered the TCP/IP group three years ago (specializing in
- > the MVS side of things), all the talk was
- > on its slow performance. That has been improved tremendously.
-
- Right. It now uses only 5 times instead of 10 times the amount of CPU
- that it should. When we converted from another vendor product to IBM
- MVS TCP/IP, our total CPU usage for all TCP/IP functions went up by a
- factor of 7-10 over the previous product.
-
- > Since that time we have added:
- > (...)
- > 5. Multitasking support for Sockets.
-
- This is what I'm complaining doesn't work, and I was told by L2 that
- this isn't going to be fixed, at least not in this release.
-
- > (...)
-
- Now, all the stuff you listed that I've replaced with (...) isn't
- worth anything to my site, and I suspect a number of other sites.
-
- > talked with Leonard Woren personally to assure him that IUCVMULT will
- > soon be reentrant, and I think that will allow him to run his
- ~~~~
- > PIE/TSO.
-
- That's half of the solution. You must anchor the table somewhere
- that's not dependent on a TCB that might go away, and you have to
- solve the problem of the IRB belonging to a TCB that may go away.
- Oh... can you define "soon"? You've been promising us privately and
- publicly for a good part of a year that fixing this is your "highest
- priority project", yet we've seen nothing. That's why we're upset.
-
- Our last communication on this subject was a proposal from you to
- front-end the IUCVMULT front-end. I pointed out why that solution
- wouldn't work, and I never heard from you again. (You also told me
- that making IUCVMULT rent wouldn't happen in the near future.) Try to
- understand why I feel the way I do.
-
- > We are always open to suggestions or constructive criticism, because
- > we improve through feedback. But venting spleen does not
- > create a healthy atmosphere for cooperation and dialogue.
-
- When I have problems which have been open for a long time, and/or
- major usability problems that shouldn't be any big deal to fix, and
- when I'm told that IBM is closing *all* of them as "submit a
- requirement," I get a bit pissed.
-
-
- Bitnet has just raised the minimum limit for mail to 1M, with a
- recommendation to support at least 3M. With MIME coming into use even
- 3M could be much too small. But IBM MVS TCP/IP out of the box doesn't
- support 3M without a zap, which is not acceptable. (Note that the
- manual states that the range for MAXMAILBYTES is 1 to 2**31-1, but
- the product doesn't actually support this. This is non-trivial to fix
- due to broken design, so IBM is not going to fix it. Do you think
- that "Submit a requirement" is a reasonable response for this??? If
- this was fixed right, it would improve the abysmal performance of
- SMTP a bit. Most customers probably aren't aware that SMTP goes
- through EOV for every piece of mail larger than one track. EOV is
- expensive. And if the mail is large, SMTP could go through EOV 15
- times! And if it's a bit bigger, SMTP abends with an SB37. The
- remote host requeues the mail for later retransmission, TCPIP auto-
- restarts SMTP, the mail is retransmitted, and SMTP abends again. And
- the only thing that the receiving site can do is change MAXMAILBYTES
- downward to whatever fits in 1+15*3 tracks. How many customers out
- there feel that this is reasonable and doesn't need to be fixed now?
-
- /Leonard
-