[Date Prev][Date Next]
[Chronological]
[Thread]
[Top]
Re: realtime networking
On Thu, 3 Sep 1998 19:34:20 Yoav wrote:
>
> I consider doing some work on realtime networking protocol over ethernet.
> (under linux of course)
>
> 1. Has anyone already done something similar ? I couldn't find anything,
> but I want to be sure before I start working.
> 2. Any interesting ideas/suggestions for the implementation ?
> Currently, I plan to implement something similar to token-ring over
> ethernet, and have an RT_sendto() function that receives the usual
> staff plus priority for the packet. Needless to say, all hosts on the
> ethernet are assumed to use this protocol, so under no circumstances
> will there be a collision. (no arps running around, etc...)
> TCP/IP + arps and the rest of the usual networking staff will work on
> this network although not being realtime, but will not affect the
> realtime communication. This will be achieved by having IP call
> RT_sendto() whenever sending a packet, and use the lowest priority.
> IP itself cannot be made realtime even if you increase its priority,
> because if the target does not exist in the local arp table, you must
> send arp, get arp reply, and only than transmit your packet. Also, tcp
> has a retransmit mechanism, which is not a good idea here. If the
> packet got lost, I dont want an auto retransmit, but an error, letting
> the application know that the packet wasnt sent on time.
>
> Any interesting thoughts about this ?
> Will such project be useful for more people, if I do it ?
Is the goal to have more deterministic timing for the messaging? Sort of a RT
network scheduler. My token-ring is more than a little rusty, but your concept
sounds similar.
There are a lot of interesting problems that would need to be solved. So many,
in fact, that it may not be practical or even posible to create "one size"
(solution) to fit all. A layered toolkit like, surprisingly enough, the OSI and
TCP/IP models with mix and match components tailored to the range of RT
messaging requirements could provide the needed structure and flexibility.
Those are my thoughts interesting or not. High speed communication is of
interest. I am several months (atleast) away from having an immediate need.
You might want to take a look a what was done in the Beowulf project. That is a
MP (massively paralell) linux. Special divers were created for high speed
networking (ethernet, passive backplane, etc.).
Regards,
david
--
David Ross
davidr@toadtech.com
Toad Technologies
"I'll be good! I will, I will !"
RT Linux Home Page: http://rtlinux.cs.nmt.edu/~rtlinux