On Fri, 2004-10-22 at 13:57, neil gardner wrote:
Well, I've never met Bruce, and I do read Aardvark most days, but I think it would be good to get someone that really _knows_ what the impact of reducing the line speed on latency will be.
We have a few dslams here for testing but they're been used for real work right now.
Anyone _know_ what the effect will be?
We can estimate some of the effects.
Back of envelope calculations (ok, I lied, I have machines that can do
this sort of stuff).
Time taken to send 1 bit at 8,000,000 bits/second: 125ns
Time taken to send 1 bit at 288,000 bits/second : 3.47us
Average SYN and SYN+ACK packet size: 60 bytes / 480 bits
Average ACK (w/out data) : 52 bytes / 416 bits
Time to send SYN/SYN+ACK @ 8Mbps : 60us
Time to send SYN/SYN+ACK @ 288Kbps: 1.67ms
Time to send ACK @ 8Mbps : 52us
Time to send ACK @ 288Kbps: 1.44ms
TCP 3-way handshake at 8Mbps : 172us
TCP 3-way handshake at 288Kbps: 4.78ms
TCP handshake increase: 4.61ms (or nearly 2800%)
I'm ignoring ATM overhead here, so these figures are going to be
somewhat higher in the real world, but it serves to give you an idea
anyway. I used the 288Kbps figure as it was used earlier in this thread.
I can understand that from Telecom's perspective, moving the
ratelimiting out to the edges of their network a more scalable way to do
things. That doesn't mean I have to like it though, I'm going to miss
leeching from JSG :-)
Regards,
Nic.
P.S. Disclaimer: I occasionally have brain farts; my maths may be wrong.
--
Nic Bellamy