At July 13, 2005 2:33 PM, Lydia Leong wrote:
> On Jul 12, 1:48pm, "Richard A. Bartle" wrote:
>> Would developers be interested in being able to observe the
>> effects of lag, latency, audio and video degradation under their
>> direct control in laboratory conditions? Or is knowing the ping
>> enough?
> There are three major components of effective transmission
> quality: Latency (your "ping"), packet loss, and jitter (the
> variability in your latency). So knowing the ping is definitely
> not enough.
Jitter is often naively modeled as Gaussian, but variance in latency
can also be driven by congestion conditions. Consider the classic
28.8kbps modem. Your ping can sit at 100ms until the moment you
push a lot of TCP data up or down the pipe. At that point queuing
delays and the modem's internal buffer will make your RTT spike to
2000ms and hold it there until the congestion goes away. Some cable
modems will simply discard packets once you've exceeded your
bandwidth allocation for the second. (I've seen them *truncate* IP
packets at 48 byte boundaries, which is highly suggestive of an ATM
backbone.)
I've heard good things about the NIST Net software package for
network emulation. Runs on a spare PC with two Ethernet cards.
http://snad.ncsl.nist.gov/itg/nistnet/ . Haven't looked at it in a
while. From the paper there:
== The set of network effects NIST Net can impose includes: packet
delay, both fixed and variable (jitter); packet reordering; packet
loss, both random and congestion-dependent; packet duplication;
and bandwidth limitations.
==
You can of course spend arbitrary amounts of money on appliances
that do the same thing. They can still be cheaper than Iridium
airtime for weekly testing, though. :-)
Jay