sl11@prism.gatech.EDU (LIEBESKIND,SUSAN H) (06/05/90)
A couple of weeks ago, I posted a request for information about Apollo's plans for support of SysVR4. In my posting, I mentioned our frustration for SysV Streams, and that a person at Apollo technical support had told us it was impossible to use SysV Streams for interprocess communication, because of the way the Apollo network interface is managed by the Aegis driver. The silence about SysVR4 was deafening - I did not hear anything at all official about support for SysVR4 in SR11 or later. However, I did get some responses back on the blanket statement about Streams. The most thorough response I received was from Jim Rees, formerly of Apollo. He had worked on Apollo's implementation of Streams, and he filled in some of the blanks for me. With his permission, I am posting excerpts from our email correspondence that which pertains to the Streams discussion, in the interests of correcting the somewhat negative impression that Tech Support gave us, and which I passed on to the net. >From: rees@citi.umich.edu (Jim Rees) >To: sl11@prism.gatech.edu (LIEBESKIND, SUSAN H) > >Streams is just a framework. Apollo does not provide any transport to go >under it, whereas sockets have at least two transports available (Unix >domain and TCP). > >The main advantage of using Streams is that if you want, you can provide >your own transport to connect up to it. You currently can't do that with >sockets, at least not easily. But if you have no interest in writing >transports and just want to move bits, you're better off with sockets. > [Regarding the statement about it being impossible to do IPC with streams..] >I don't believe this, but it would be difficult for me to produce a >counter-example. Without knowing more, I would guess that your >correspondent couldn't figure out how to share a network device between >Domain/OS (aka Aegis) and Streams. > >As far as I know, Apollo's implementation of Streams works as well as >anyone's, and the trouble you had was with AT&T's specification of Streams >and the lack of device drivers rather than with Apollo's implementation. By >the way, I don't know of any Streams implementation that comes complete with >transport layers or device drivers -- there were none at the time we did our >implementation, but there may be some now. > My thanks to those other folks who did respond to me with information on streams, Jinfu Chen of Motorola, and Larry Allen of OSF. Hope this helps other folks out there wrestling with streams vs. sockets issues. Susan Liebeskind