Re: one phenomenon on ATM routing

[Jon Crowcroft <[e-mail]>]

 >We have an ATM network which is configured by two ATM switches(buffalo 
 >and cowboy). Now, I just connect two workstations(SUNSparc5 and SGI Indy) 
 >to buffalo. When I traceroute from buffalo to every ATM network element, 
 >I find a strange phenomenon. The first packet of the first time I 
 >traceroute on needs much more round trip time. Does that mean ATM need more 
 >time to negotiate, comparing to ethernet? Is that reasonable? The following 
 >info is about the traceroute:
 >to cowboy:
 >the first time: 402ms, 3ms, 3ms
 >the second time: 4ms, 3ms, 3ms


right - you wouldn't want to be doign transaction processing over
such as service would you? :-)

its the VC setup time (though why its so large, i don't know - we
don;''t see anything like .4 to .9 secs on the switches we use, more
like .1 secs)

whose switches are you using, and what signalling do they
support....?

also, if you are using them between routers, why don't you just
configure PVCs (while we await ABR standardisation)?

but imagine you were a bank with a big ATM (Automatic Teller Machine)
account server, processing 10,000s of atm (asynchronous transfer mode) 
calls per second from remote ATMs - you'd be worried

i suppose you'd configure a VC from every ATM over the atm to your
server, but if they traversed a pay-for-use net, you'd get concerend
over call-time charges, wouldn't you? but then i guess, call-time
charges will just have to go:-)

jon