I've been pondering what aspects of a residential broadband connection that would be worthwhile in testing, which would also be some kind of incentive for ISPs to start doing "better".
Some things that comes to mind:
speed
latency to some points geographically near the user
MTU of the connection
If PMTUD works or not
queueing (FIFO or something "better")
antispoofing (BCP38) compliance
filtering (IPv6 transition protocols for instance, lots more possible)
buffer depth ingress/egress
ECN
ISP provided DNS resolver properties (DNSSEC, EDNS etc)
I'm sure there are lots more, and this could probably not be done using a web browser driven application, but instead would have to be an application, thus harder to get people to use generally.
Any work being done in this area already that someone can point to? I'd also like to use some of this tech to do "web server tests", especially when it comes to PMTUD working, especially when for a IPv6 world it would be nice to have an easily available testing suite for these basic mechanisms.
[on residential broadband connections]
Mikael Abrahamsson writes:
Some things that comes to mind:
speed
latency to some points geographically near the user
MTU of the connection
If PMTUD works or not
queueing (FIFO or something "better")
antispoofing (BCP38) compliance
filtering (IPv6 transition protocols for instance, lots more possible)
buffer depth ingress/egress
ECN
ISP provided DNS resolver properties (DNSSEC, EDNS etc)
That's an excellent start. I would add
* availability of global IP addresses (0-n)
* ability to connect to "unusual" ports (falls under "filtering")
* ability to accept connections
* interception of common TCP ports such as 80 and 25
* transparency for various header fields (addresses & ports¸ DSCP...)
* rate-limits for specific protocols (ICMP, BitTorrent...)
* latency and throughput for some popular sites/resources, including
those using CDNs, at various times of day/week
...and of course...
* availability of IPv6
I'm sure there are lots more, and this could probably not be done
using a web browser driven application, but instead would have to be
an application, thus harder to get people to use generally.
Any work being done in this area already that someone can point to?
Check out M-Lab - http://www.measurementlab.net/
Offline suggestions I have received also includes http://netalyzr.icsi.berkeley.edu/ which seems to have some of the test suggestions as well.
Hi,
I'll toss in that I2 and GEANT have been developing the PerfSONAR toolset.
http://www.perfsonar.net/
Regards,
Chris