Not logged in
Log in now
Create an account
Subscribe to LWN
LWN.net Weekly Edition for December 5, 2013
Deadline scheduling: coming soon?
LWN.net Weekly Edition for November 27, 2013
ACPI for ARM?
LWN.net Weekly Edition for November 21, 2013
Connecting on the QUIC
Posted Jul 17, 2013 17:32 UTC (Wed) by felixfix (subscriber, #242)
Posted Jul 17, 2013 18:05 UTC (Wed) by renox (subscriber, #23785)
I think that he knows that, but if protocols could at least avoid wasting RTT (like TCP does), it would be a progress.
Posted Jul 20, 2013 3:09 UTC (Sat) by giraffedata (subscriber, #1954)
Likewise, I question the Chromium blog author's motivation of the project to reduce round trips by saying the speed of light is constant. I routinely work across a link which is 2000 km as the crow flies. The constant speed of light limits me to 13 ms round trip, but I see about 100 ms ping time. There are apparently some significant variables involved in the length of a round trip.
Posted Jul 20, 2013 4:08 UTC (Sat) by felixfix (subscriber, #242)
Ping times range from 1.3 to 2.0 seconds. I suspect the satellite and ground station are buffering as much as possible to maximize payload per packet, and that communication between the satellites and modem / ground stations is some very packed non-TCP protocol.
Posted Jul 20, 2013 16:14 UTC (Sat) by giraffedata (subscriber, #1954)
The signal goes up, down, turns around, up, and down again. It's very roughly half a second round trip.
Ah, I forgot you're pinging something else on Earth, not the satellite itself. So the physical constant C's limitation is, by my calculation, .5-.6 seconds per ping.
Posted Jul 20, 2013 7:28 UTC (Sat) by Cyberax (✭ supporter ✭, #52523)
Also, on very short distances your latency can be dominated by the fixed switching latency.
Posted Jul 20, 2013 10:21 UTC (Sat) by hummassa (subscriber, #307)
ping time = 2 * (distance * c + sum(latency-per-hop))
is a reasonable approximation. Latency-per-hop is usually in the 5-15ms range for stupid or switched hops and anywere in the 25-200ms for routed/firewalled hops. I suppose a special protocol-changing hop could cost even more.
Posted Jul 20, 2013 16:06 UTC (Sat) by giraffedata (subscriber, #1954)
You might want to check the real physical path of the fiber... Also, on very short distances your latency can be dominated by the fixed switching latency.
Yes, those are two of the variables I'm talking about, showing that the constant speed of light doesn't by itself mean the only way to reduce latency is to reduce the number of trips.
Copyright © 2013, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds