home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!gatech!paladin.american.edu!auvm!UCLAMVS.BITNET!CSYSMAS
- Message-ID: <IBMTCP-L%92122218145799@PUCC.PRINCETON.EDU>
- Newsgroups: bit.listserv.ibmtcp-l
- Date: Tue, 22 Dec 1992 15:12:00 PST
- Sender: IBM TCP/IP List <IBMTCP-L@PUCC.BITNET>
- From: Michael Stein <CSYSMAS@UCLAMVS.BITNET>
- Subject: Usage of vector facility for TCP/IP?
- Lines: 24
-
- > Since soon a major X-terminal wave will hit us I am looking
- > oncemore for ways to reduce the CPU consumption of TCP/IP.
-
- Reduce the number of instructions in TCP/IP. It's way too high...
-
- > Is there a way to use the vector processors on the mainframe for
- > the TCP/IP protocol stack and/or the X-Window system?
-
- It might be possible to compute the TCP checksum using the vector
- hardware for large packets and thereby save some CPU time.
- Unfortunately, since TCP doesn't spend all that much time of the
- CPU it uses computing the checksum that wouldn't save all that
- much CPU.
-
- > I guess this is a vital question for IBM. Up to now more than
- > 90% of our CPU has gone into batch. If this batch load would be
- > severely diminished by 200+ (concurrent) X-terminals running
- > X-Window based graphic applications nobody might listen anymore
- > to pro-mainframe arguments in a scientific environment.
-
- Yup.
-
- "You're selling me the fastest general available computer on the
- planet and it connects to my network via this straw?"...
-