home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: vmsnet.networks.tcp-ip.multinet
- Path: sparky!uunet!mcsun!sunic!sejnet.sunet.se!eric
- From: eric@sejnet.sunet.se (Eric Thomas)
- Subject: Re: Do UCX and Multinet support nonblocking I/O?
- Message-ID: <1992Aug31.024421.1@sejnet.sunet.se>
- Lines: 28
- Sender: news@sunic.sunet.se
- Reply-To: ERIC@SEARN.SUNET.SE
- Organization: SUNET, Stockholm, Sweden
- References: <dank.715215744@blacks>
- Date: Mon, 31 Aug 1992 02:44:21 GMT
-
- In article <dank.715215744@blacks>, dank@blacks.jpl.nasa.gov (Daniel R. Kegel) writes:
- > Hi,
- > I want to port to VMS a Unix program which initiates several
- > connections to TCP servers and uses nonblocking I/O
- > to manage all the connections simultaneously.
-
- Porting unix socket-based code to VMS is easy (at least with Multinet), it
- mostly compiles the first time and making it work is just a matter of replacing
- the many system("rm logfile") with more appropriate calls. Finding out that it
- doesn't quite perform as well as similar native VMS utilities is the bad
- surprise. Realizing that the VMS-ported server isn't slower than the native
- unix one only takes a minute or so, and deciding to do something about it is a
- matter of seconds if, like me, you cannot stand things which are unnecessarily
- slow.
-
- Changing the server to use asynchronous calls everywhere is easy and solves the
- problem admirably (again with Multinet, and assuming you are familiar with
- AST's). The difficulty is convincing unix people who are impressed with the
- performance of your server on that despicable little 2 mipser and would like a
- copy of your code that this cannot be done under unix. Just sending them the
- code won't work, because they will come back 10ms later with questions about
- all this IO$_ crap. Explaining takes time and usually results in religious
- statements about why Things should not be done that way, which is precisely why
- it can't be done with unix. Eventually you will wish you had just put up with
- the slow response time instead of trying to prove to the world that things
- don't have to be that slow - my advice ;-)
-
- Eric
-