>>>>> "Guilherme" == Guilherme Nelson F De Souza <[EMAIL PROTECTED]> writes:
>From: "Dresner, Norman A." <[EMAIL PROTECTED]>
>> Token-ring networks are inherently non-deterministic (within
>> sloppy limits, I suppose, depending on the specific network
>> protocols governing how long messages can be and how many messages
>> a station can transmit before passing the token to the next
>> station),
Guilherme> Again, I may be wrong (have forgotten my network protocol
Guilherme> class), but as far as I remember, token-ring networks ARE
Guilherme> deterministic. Every message can be guaranteed
Guilherme> (probability one) to be delivered within a fixed number of
Guilherme> cycles (passing of the token).
Your last sentence is correct but what precedes it is not. The reason
is that there is no bound on how long it takes for the token to come
around.
In the ERROR FREE case, token rings have a tight upper bound on how
long it takes to transmit. By the way, that bound is MUCH longer than
you would expect. For example, with FDDI it is TTRT (often 8 ms, but
it can be as high as 165 ms) times the number of stations on the ring.
With 802.4, ditto. With 802.5 it's much longer. (802.5 and FDDI have
*nothing* in common other than that they both use tokens.) So yes,
you have a hard bound, but that bound may well be way over a
second... For an FDDI of max size with max legal TTRT, the bound is
almost 3 minutes. :-) That number is so much larger than the typical
transmit time that Norm's comment is valid for practical purposes even
though it's not 100% accurate in theory.
On the other hand, in the presence of errors, even that very large
bound goes away because now you have to ask how long it takes to
recover the token once it's lost. That process is bounded for FDDI in
the error free case, but not in the case where you have some sort of
problem. I don't know if it's well bounded at all in 802.5.
I have a very simple attitude on this: (a) just about any network has
useful real time properties in the absence of errors and with
reasonable design (Ethernet in particular included). (b) NO network
whatsoever has or ever will have real time bounds all the time, since
there can never be such a thing as an error free network, nor for that
matter an error free bus or an error free CPU. (c) if you know what
the acceptable failure rate is, you can design a network that will
satisfy that requirement, and you'll find there are many ways to do
that. (d) if you do *not* know the acceptable failure rate, or you
think it is zero, you will end up with a useable implementation only
by blind luck.
paul
--- [rtl] ---
To unsubscribe:
echo "unsubscribe rtl" | mail [EMAIL PROTECTED] OR
echo "unsubscribe rtl <Your_email>" | mail [EMAIL PROTECTED]
----
For more information on Real-Time Linux see:
http://www.rtlinux.org/~rtlinux/
