adam@its63b.ed.ac.uk (ERCF02 Adam Hamilton) (03/13/87)
As the various Ethernets at Edinburgh University get busier, some people start to worry that they are running out of capacity. Occasionally they ask me "when do we NEED to split up the load". ( The usual suggestion is bridges). I have no feel for what proportion of the Ethernet bandwidth can be used before performance suffers. Would someone out there like to offer me the benefit of their experience. I will certainly summarise e-mail responses but I suspect that a direct posting by a net administrator who has been through this would be better. Our monitor shows that typical day-time use is about 10% of the maximum bandwidth over 1 minute periods with 1 second peaks of up to 50%. This is our CS department, and they want to add another 15 or so Suns. Another department (A.I. no surprise) shows even more use than this with bursts of 10-15% (over 1 second). The main heavy users seem to be diskless Suns - a node and its server can use about 5% in bursts. My own view is that if the traffic made a 1-second paging time change to a 2-second paging time that would be acceptable; if it made other traffic substantially worse, that would not. (Yes, I know this is very woolly but I don't know any better.) OK, the question is HOW BUSY CAN AN ETHERNET BE before it NEEDS to be split? Adam Hamilton !seismo!mcvax!ukc!ed.its63b!adam I think P.S. Make that 30% steady average and up to 70% peaks. Now I am worried :-!