Re: comment on intel i-d on h323 and internet

[Jon Crowcroft <[e-mail]>]

this is a seperate but useful point......

but yes,. why h323 can't just use either TCP conenctio ncompletion as
implict q931 setup completion, or even use a multicast control
protocol (actually they are working on that now) was always beyond
me.......ditto for the H245, at least that could remove 2 RTTs or even
3 or 4 if you remove TP0 mapping....

[we've been here before a few times, havnt we!!!]

but my main concern was about the multipoint controller - the order of
arrival of conference control application messages is non determined,
where as with a fixed RTT yo uhave in an ISDN network (viz it must be
or BONDing wouldn;t work:-), yo ucan do one time RTT estiamtion ,and
then create a perfect ordering with a trivial application protocol,
which simply wont work over a star of TCP conenctions....so the
service, e.g. for distributed whiteboards, will NOT have the same
semantics...

oh well...

 >Last time I counted, doing RAS + q931 + H245 implied :
 >
 >1 RTT for RAS,
 >3 RTT for q931:
 >        1 for the TCP SYN,
 >        1 for the TP0 SYN in RFC1006
 >        1 for the q931 call set up
 >4 RTT for h245:
 >        1 for the TCP SYN
 >        1 for TP0
 >        1 for the capacity exchange
 >        1 for the channel open
 >which implies 8 RTT before you set up, say, a telephone call.  The
 >current Internet RTT for a long distance transmission is at best 100 ms,
 >more likely 200.  So, we are speaking of 0.8 to 1.6 seconds to set up
 >the call.  If you have any transmission error, you have to add 4 seconds,
 >i.e. the initial timer estimate of TCP.  and we can certainly expect
 >quite a few errors on long distance transmissions...


yep,
cheers

 jon