The idea is nice, using the net like a bubble memory, but capacy is rather small, since any delayed loop storage is determinated by bandwidth and trip time. Let's just assume we have a 1 Mbps connections (full dupelx), so we can send (and recive) 128 KBytes/s. If we now add a delay time for roundtrip of eight seconds (prety high - 1s would be more realistic), we just get along with 1 Mega- byte of storage. As you see, the limit itn't the net (and the possible net resources to use), but rather the bandwidth of our connection. And dont tell me you would open more 'storeage connections' to expand - we have already assumed the full bandwidth of our 'memory interface' - you can't put more data in. Ethans solution of derefered methods is more usable, since it would enhance the latency time - even if we assume a pure memory based solution (no mail buffer on HD), we can come up with several minutes of delay (but in fact, all mail servers will use temp HD storage - but hey, /temp isn't considered as disk - it is just always to small :). Gruss H. -- Stimm gegen SPAM: http://www.politik-digital.de/spam/de/ Vote against SPAM: http://www.politik-digital.de/spam/en/ Votez contre le SPAM: http://www.politik-digital.de/spam/fr/ Ich denke, also bin ich, also gut HRK

Yes and No - maybe we should clarify terms. a) round trip time (rt)- the time uses until one pice of information (packet) is returned - for calculation it is enough to use an (hypothetical) average time b) access time (at) - access time defines the time needed to access a specific data unit within the storage - the access time has no influence on storage capacity. c) storage capacity (s) - maximum amount of data to be stored at a given moment in time d) bandwidth (f) - the amount of data that can be send/recived within a given time (at this definition level bandwidth and storage capacity is equal, but to help calculation we should specify given time to be 1s at all calculations) These terms should be incooperated as following: I) s = rt * f - since you can only put a continous data stream of f data units for one period of rt until you need _all_ f for refresh II) at(min) = 0 at(avg) = rt/2 at(max) = rt or 0 <= at <=rt Since rt and f are given (exterior) values, they have to be assumed. Looking at these numbers, you might come to the conclusion that s is the only number needed to describe the system and s is strict dependant to rt and f. (please excuse my missing English on mathematical terms). Followin this, we may insert your names (as I have understood them): latency ^= round trip time (rt) write speed ^= bandwidth (f) read speed ^= access time (at) See the confusion ? even then, 8 seconds is _pretty_tough_. given an average switching time of .2 seconds it needs 40 hops - transmission time can be ignored at this stage, since we can't name the data amount - and transmission time is amount dependant. As soon as you like to have longer delay times, you _need_ explicite storage inbetween (remember, even a satalite linke is just .2 seconds round trip) - and you won't have a lot on your chain - satelite lines are a NoNo for (modern) hige speed interactive connections. Gruss H. And for the naysayers, I assume this on topic, since it on topic for the basics of delay dependant storage. -- Der Kopf ist auch nur ein Auswuchs wie der kleine Zeh. H.Achternbusch