alfs messaging (IPC)

Jason Gurtz jason at tommyk.com
Thu Feb 15 09:18:30 PST 2001



> wow this blew by everyone and no one had
> any comments besides jay? So then go forward
> with it?

hehe, maybe they were sleeping...  ;)

> > > Again the alfs_app would just do whatever with %messages. Since
> > > the frontend
> > > is the one who sent the profile, they can receive these %messages
> > > and process
> > > them.
> >
> > I think it would be a good thing if the messaging had tcp like QOS
> > guarantees in that BE sends message with MD5Sum and the FE gets
> that checks
> > the MD5Sum and if OK, sends back to the BE, "i got this OK;
> readyForSend()"
> > else "the net/commChan is FUBAR; reSendMsg()"  (or thereabouts behaviour
> > wise  ;)
> > In theory, this would really only be kinda inportent with error
> and input
> > passing, and could really get away with not having check sum
> and only using
> > unique message ID's and just answering back I recieved message with
> > ID#:<foo>
>
>
> ok you have confused me. Can you explain this again. We are working
> in a tcp enviroment I thought, not udp. How are these messages like
> udp and would need your additional checksum? I am just confused on
> why....

  I was thinking in the framework of sometimes the frontend and backend
might not be on the same machine.  If they are always togeather, then it
doesn't make any sense of what i was talking about.
  The underlying protcol (tcp) has it's own built in checksums and all that,
but the alfs system doesn't know about that.  I mean, what if behavior like
what happens with my NAT machine at home happens. (the behavior is that
select http conections seem not to work compleatly i.e. i can't view certain
web sites)  Sure, this is probably a (minor)misconfiguration somewhere, but
it sure would be a show stopper with alfs.
  So, basicly the checksum, or simple msg id checking would be another layer
at the app level, and a further check against loss on the network that might
not get taken care of.  Now that i'm thinking of this again, 'simple msg ID
checking' seems reasonable enough as a message can still be interpretted if
one character of it is mangled.

>
>
> > Should their be a difference between warning and error like
> with compiling
> > or is simple pass/fail good enough?
>
> to me a warning is a message, not an error/input....

Yeah, seems right  :)

~jay

--
+------------------+
| Jason at tommyk.com |
+------------------+






More information about the alfs-discuss mailing list