lerner@isi-vaxa.arpa (Mitchell Lerner) (05/06/86)
>Does anyone have any information or experience regarding the general >health of stream-mode, unix-domain sockets under Release 3.0 of SUN >UNIX? We are helping some folks port an application from a BSD 4.2 >VAX 750 to our SUN-3/160. The application appears to run fine on the >750, but on the SUN, the processes seem to have problems trying to >rendevous at each others sockets. Occasionally everyone comes up >and all appears to run fine, but usually the rendevous fails. Well, sombody correct me if i am wrong, but, when 4.2 first was released Unix domain sockets were flakey, much in the way that you describe above. I was told that Dr. Bill Joy himself said, somthing to the effect: "Unix domain sockets were just not a good Idea." and " Dont use Unix domain because It doesnt work.". Even for Intrahost (same host) IPC I've used the Internet domain and had no problem. The kernal can tell that the connection is not remote and doesnt send the data out the e.c/LAN (incurring overhead). I'm not shure that this is what you wanted to know but ... for what it's worth ...
mark@umcp-cs.UUCP (Mark Weiser) (05/08/86)
In article <596@brl-smoke.ARPA> lerner@isi-vaxa.arpa (Mitchell Lerner) writes: >I was told that Dr. Bill Joy himself said, somthing to the effect: ... Is B.J. a Dr. now? -mark -- Spoken: Mark Weiser ARPA: mark@maryland Phone: +1-301-454-7817 CSNet: mark@umcp-cs UUCP: {seismo,allegra}!umcp-cs!mark USPS: Computer Science Dept., University of Maryland, College Park, MD 20742
lerner@isi-vaxa.arpa (Mitchell Lerner) (05/09/86)
I've found it interesting that out of six responses to my jounal submittion, six of them were responses only concerning if indeed bill joy has a Ph.D.. hmmmmm, considering the size of his following and the current glamor of computing in the media, maybe this is good matieral for the National Star or the Enquirer? Bill, Are there any juicy tid-bits about your background that might interest inquiring minds (like our's on the net)? You might become the first computer scientist to make the "Enquirer set" or Dell books. "Unix domain sockets" could be a house-hold phrase just like "theory of general relativity".
bbarnett@factron.uucp (Bruce Barnett) (05/12/86)
In article <596@brl-smoke.ARPA> lerner@isi-vaxa.arpa (Mitchell Lerner) writes: >>Does anyone have any information or experience regarding the general >>health of stream-mode, unix-domain sockets under Release 3.0 of SUN >>UNIX? >Well, sombody correct me if i am wrong, but, when 4.2 first was released >Unix domain sockets were flakey, much in the way that you describe above. >I was told that Dr. Bill Joy himself said, somthing to the effect: "Unix >domain sockets were just not a good Idea." and " Dont use Unix domain because >It doesnt work.". > >Even for Intrahost (same host) IPC I've used the Internet domain and had >no problem. The kernal can tell that the connection is not remote and >doesnt send the data out the e.c/LAN (incurring overhead). Doesn't this work only if you run ifconfig(8)? Then, assuming a stand-alone machine, you either get an error `no carrier' every thirty seconds (or so) or you have to put a loop-back plug on the ethernet connecter on the back of the Sun. My question is: Is there a single method of using sockets on both stand-alone and networked machines? Or do we have to use two different portions of code for the two configurations? How do people handle this difference? As an aside, if Unix Domain sockets are so flakey, how do pipes (which are implemented as sockets) work without (apparent) problems? Bruce Barnett (518) 783 3516 seismo!rochester!steinmetz!factron!bbarnett spar!factron!bbarnett
jas@rtech.UUCP (Jim Shankland) (05/12/86)
> Even for Intrahost (same host) IPC I've used the Internet domain and had > no problem. The kernal [sic] can tell that the connection is not remote and > doesnt send the data out the e.c/LAN (incurring overhead). (plus ca change...) But I've recently had cause to discover that using the Internet domain intra-host is still twice as slow as using the Unix domain, apparently because at least some layers of the stream protocols are still being used (checksumming, etc.). Jim Shankland ihnp4!cpsc6a!\ rtech!jas ucbvax!mtxinu!/
jdb@mordor.ARPA (John Bruner) (05/15/86)
Aside from issues of speed, UNIX domain sockets provide a capability (no pun intended) that Internet domain sockets do not. It is possible to pass file descriptors between processes using UNIX domain datagrams. -- John Bruner (S-1 Project, Lawrence Livermore National Laboratory) MILNET: jdb@mordor [jdb@s1-c.ARPA] (415) 422-0758 UUCP: ...!ucbvax!dual!mordor!jdb ...!seismo!mordor!jdb
rbj@icst-cmr.arpa (Root Boy Jim) (05/19/86)
Bill, Are there any juicy tid-bits about your background ... Here is a haiku I found while looking thru /usr/src/ucb/ex/expreserve.c: /* * people making love * never exactly the same * just like a snowflake */ You find things in the strangest places. (Root Boy) Jim Cottrell <rbj@cmr> "One man gathers what another man spills"
brett@wjvax.UUCP (05/19/86)
In article <6928@mordor.ARPA> jdb@mordor.UUCP (John Bruner) writes: >Aside from issues of speed, UNIX domain sockets provide a capability ... >to pass file descriptors between processes using UNIX domain datagrams. I am curious -- how? In addition, I have a general question. It looks to me like the name to which a UNIX domain socket may be bound is limited to 13 or fewer characters (at least in our implementation - VAX 750, Mt Xinu 4.2 BSD) in the struct sockaddr structure. UNIX domain socket names are located in the filesystem, and this 13 character name length limit seems severely to restrict where in the filesystem a socket name may reside. How do other people manage this restriction? Is there such a restriction? If so, where do people typically put socket names -- in /etc? Please respond by E-Mail; I will summarize. Thanks (in advance) ------------- Brett Galloway {pesnta,twg,ios,qubix,turtlevax,tymix,vecpyr,certes,isi}!wjvax!brett