pom@under.s1.gov (Peter) (09/11/87)
Well, well This is really the right discussion group for me. One does not get smothered long tracts. I recall somewhat irritated reaction to my sweeping question on merits of Synchronous vs Asynchronous Protocols ( SP vs AP ). As I do not want to be left Holding the Token for ever, I will respond to that comment: The answer that Ethernet is both SP and AP is not correct. Ethernet does 'exponetial backing' after collision - which means that PROBABILITY of sending a packet is determined by a formula. Neverthless, such protocol remains fully asynchronous. Better example of protocol, which has both ( AP and SP) is Hyperchannel -- which was described in a very imaginative way in Mitre book on LANs . Hyperchannel enters "Free for all" mode after (fixed) period of silence. In that mode, any station can send - without waiting(=AP). In case of collision, nodes enter into 'slotted mode' in which each each station is assigned a 'fixed time interval' or 'slot' in which it can send a packet (SP). My question attempted to draw attention to that universal parameter, ( which Tanenbaum calls 'a' and ) which is a (dimesionless) ratio of two time-intervals 1) rate of transmition * length of packet 2) Speed of propagation * length of the bus General dependence of throughput on a T[a] says that in 'heavy traffic' you prefer SP and in 'light traffic' you prefer AP. This universal tendency is seen even on (very simple example of a ) traffic lights: In low traffic, the light should revert to 4-way stop ( red blinking) and only AFTER FIRST COLLISION, they should revert to usual slots ... Did I missed something? It looked so logical, when I started.. Anyway, all those with first name in [a...k] should now post a statement of their interest [ LANs, VANs, ...??? ??? ] Yours Peter || pom alias pom@under.s1.gov || pom@s1-under.UUCP
rpw3@amdcad.AMD.COM (Rob Warnock) (09/17/87)
In article <3819@sdcsvax.UCSD.EDU> pom@under.s1.gov (Peter) writes: +--------------- | Anyway, all those with first name in [a...k] should now post a statement of | their interest [ LANs, VANs, ...??? ??? ] +--------------- Hmmm.... obviously an attempt to get us to use a version of the "Urn" algorithm, often said to be near-optimal under mixed loads. Unfortunately, Urn algorithms require universal acceptance of the rules, which I seem to have just violated... ;-} ;-} Fortunately, except for considerations of disk space, USENET is not a contention protocol, but queued (assuming "inews" locks work correctly!)... In a more serious vein, Urn algorithms *do* require universal knowledge of the "state" of the algorithm (in this case, the state is that we are offering permission to transmit only to "[a-k]*"), and for this reason have some difficulty in cases such as packet radio when the packet loss rate is high. You then have to have various heuristics to "guess" at the global state of the algorithm, so you don't speak out of turn. Rob Warnock Systems Architecture Consultant UUCP: {amdcad,fortune,sun,attmail}!redwood!rpw3 ATTmail: !rpw3 DDD: (415)572-2607 USPS: 627 26th Ave, San Mateo, CA 94403