home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sources.d
- Path: sparky!uunet!paladin.american.edu!howland.reston.ans.net!wupost!cs.utexas.edu!geraldo.cc.utexas.edu!portal.austin.ibm.com!awdprime.austin.ibm.com!linas
- From: linas@austin.ibm.com (Linas Vepstas)
- Subject: Distributed Processing Package
- Originator: linas@boardhead.austin.ibm.com
- Sender: news@austin.ibm.com (News id)
- Message-ID: <C0Gqtn.16D0@austin.ibm.com>
- Date: Thu, 7 Jan 1993 02:56:11 GMT
- Reply-To: linas@boardhead.austin.ibm.com
- Organization: IBM Graphics Systems
- Keywords: MP distributed multi-tasking multi-processing
- Lines: 93
-
-
-
- Can someone out there help me find a permanent home for some
- distributed multi-processing code? I have some source I wish to
- put into the public domain -- somewhere where people can find it
- & get it. An anon-ftp site would be great!
-
- Description
- -----------
- The package is called "MAY" and implements a basic
- message-passing distributed-processing API. It allows you do
- develop programs that execute concurrently on a number of
- machines on a network.
-
- The API supports basic distributed services & a processor farm.
- The basic services provide a very simple, very easy-to-use
- interface to TCP/IP together with a remote server daemon that
- allows processes to be started remotely. The processor farm
- provides an equally simple API to allow repetitive work to be
- "farmed out" to a number of network nodes. It has a bunch of
- high tech built into it -- high & low watermark FIFO's,
- queue-empty alarms, load-balancing, etc. to make sure that
- each network node is busy crunching on something, and
- gaurenteeing the shortest possible real-time execution.
-
- The basic services include a "create" routine -- sort of like
- a network fork, or rexec -- it starts up program /u/joe/xyz
- on machine abc.whatever.domain. In addition, it establishes
- a network connection, i.e. a socket, between the parent and
- the child, so that the two can pass messages between one another.
- In addition, there is a "send" routine, a "recevie" routine, a
- "terminate" routine, and a routine that the child process can
- invoke to get it's parents address. The basic services are
- implemented in both TCP/IP and in UDP/IP.
-
- The package comes with documentation & examples. The processor
- farm examples include a fractal drawer, and a ray-tracer
- (a modified version of rayshade), displaying into an X11 window
- in more-or-less real time (depending on your network bandwidth &
- processor speeds). The package compiles & runs on IBM RT's,
- RS/6000's and on SGI boxes. The code is free and unencombered by
- funky copyrights.
-
- Pro's, Con's & Caveat's
- -----------------------
- The package was developed to overcome what seemed to be a basic
- limitation with RPC's -- the inability to perform asynchornous,
- distributed processing. Prior to development, the Apollo NCS
- system was examined, but seemed to be excessively complex.
-
- MAY does not support NIDL -- although it probably should. If
- the messages you are passing around contain char's, short's
- long's & double's, plan ahead to do some word-alignment/byte
- swapping accross different CPU architectures.
-
- MAY allows you to launch processes remotely, but does not put
- them there. You have to get them there first, by hand -- I used
- ftp, and later, NFS.
-
- To perform a MayCreate() (i.e. a remote fork/exec) on a remote
- machine, you must have the MAY daemon running on the remote
- machine. I did this by hacking the rc, and later, the inetd
- files on the remote machines, so that a reboot didn't force you
- to go off & restart the daemons by hand.
-
- MAY does not stop to ask you for a password -- it doesn't do ANY
- security verification stuff. And anyone using it can run most
- anything executable -- so its a real security hole if you want
- to use it in an unfreindly environment. Adding security
- features could make MAY broadly acceptable to the general
- computing public.
-
- ----------
-
- If anyone is interested, please let me know. I'm tired of
- watching my slick technology gather dust & grow old on my
- machine.
-
- --- linas
-
- ^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^
- Linas Vepstas
- Phone: 1-(512)-838-1116
- internet: linas@innerdoor.austin.ibm.com
- Home addr: 1518 Enfield Road, Austin TX, 78703-3424
- ^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^
-
-
- --
- ^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^
- You are in a twisty little maze of standards, all conflicting.
-
- ^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^
-