Subject: GSM/GPRS/EDGE delays [OT] [was: ...tail recursion elmination... ]
From: rpw3@rpw3.org (Rob Warnock)
Date: Tue, 15 Jun 2004 20:10:37 -0500
Newsgroups: comp.lang.lisp
Message-ID: <FsidnQII3d2QA1LdRVn-vw@speakeasy.net>
[This is *way* off-topic, but since he asked...]

Svein Ove Aas  <svein+usenet01@brage.info> wrote:
+---------------
| Rob Warnock wrote:
| > [1] AT&T Wireless's GSM/GPRS/EDGE cellular data service, which,
| >     while it has pretty good bandwidth [3-7 times dialup modem],
| >     has *terrible* round-trip latency:
| > [high latency ping]
| 
| Hang on, isn't that the same link you use to talk with?
+---------------

Yes. But voice calls pay the call-setup delay [see below] only once
per call, then they are billed continuously for the minutes that the
call stays connected.

+---------------
| Where's the latency, then?
| I'd be most interested in a traceroute.
+---------------

Almost all of the latency is between the cellular modem and the first
"service processor" [which is still behind the NAT gateway, before the
data even gets to the Internet]. It appears that, being a "best-efforts"
service, data is carried on an "as-available" basis, which in practice
seems to mean that they're using "fast circuit-switching" -- that is,
a circuit-switched call from the modem to the first IP service processor
is set up and torn down for *each* burst of IP packets. That means that
once-a-second things like "ping" (ICMP ECHO_REQUEST) and "traceroute"
(UDP to a random port) suffer a full call-setup delay per packet.

Traceroute from the cellular modem side [first few only]:
traceroute to fast.rpw3.org (66.93.131.53), 64 hops max, 44 byte packets
 1  10.251.1.26 (10.251.1.26)  974.123 ms  609.204 ms  629.911 ms
 2  10.251.0.1 (10.251.0.1)  599.648 ms  609.683 ms  629.716 ms
 3  10.250.1.2 (10.250.1.2)  619.403 ms  609.467 ms  629.447 ms
 4  allntxgdmzrt01-f4-0-0.network.attwireless.net (209.183.48.2)  599.935 ms  618.991 ms  599.909 ms
 5  allntxgisprt01-f1-1-0.network.attwireless.net (172.16.1.102)  599.503 ms  769.224 ms  639.480 ms
 6  allntxgisprt02-f4-1-0.network.attwireless.net (172.16.1.226)  409.877 ms  619.285 ms  599.623 ms
 7  se-2-0-1.hsa1.Dallas1.Level3.net (209.245.240.21)  499.499 ms  619.134 ms  599.913 ms
 ...

Ping from the other side (my server) back to the host listed in #7 above:
PING se-2-0-1.hsa1.Dallas1.Level3.net (209.245.240.21): 56 data bytes
64 bytes from 209.245.240.21: icmp_seq=0 ttl=250 time=54.419 ms
64 bytes from 209.245.240.21: icmp_seq=1 ttl=250 time=52.418 ms
64 bytes from 209.245.240.21: icmp_seq=2 ttl=250 time=53.107 ms
...

As I said, almost *all* of the delay is between the cellular modem
and the first ATTWS-internal IP service processor...


-Rob

-----
Rob Warnock			<rpw3@rpw3.org>
627 26th Avenue			<URL:http://rpw3.org/>
San Mateo, CA 94403		(650)572-2607