Hi,
Does anybody have more or less recent data on the average, median and maximum
diameter (ip hop count) of the Internet?
First, to give some hints regarding the initial question: A year ago I
did some analysis based on Caida's routed /24 topology data set
(https://www.caida.org/data/active/ipv4_routed_24_topology_dataset.xml)
for data at the beginning of Jan. 2015. Its not using all available data
but rather only traceroutes that reached the destination. The attached
figure shows violin plots for each day - vertically, they show the
distribution of Hopcounts (looks a little weird due to Hopcounts only
beeing Integers).
However, please keep in mind:
i.) the data set is collected from only few physical locations but has
traceroutes towards every routed /24 prefix.
ii.) only a subset of the entire data set is shown.
iii.) its from 2015.
iv.) there is a good chance that it is not representative for the
"entire" Internet.
Secondly, regarding the ongoing discussion:
+1 for Tim's answer. IMHO, neither AS paths nor IP paths, in general,
are reliable proxies for e.g. latency or physical distance. In addition,
keep in mind that we are only able to observe a certain part of the
Internet and thus it's hard to make claims about the "entire" Internet.
best regards,
Lars
Am 22.11.18 um 10:55 schrieb t...@pelican.org:
On Thursday, 22 November, 2018 05:30, "William Herrin"
<b...@herrin.us> said:
> Good question! It matters because a little over two decades ago we had
> some angst as equipment configured to emit a TTL of 32 stopped being
> able to reach everybody. Today we have a lot of equipment configured
> to emit a TTL of 64. It's the default in Linux, for example. Are we
> getting close to the limit where that will cause problems? How close?
If it's hop-count that's interesting, I think that raises a question
on the potential for a sudden large change in the answer, potentially
with unforeseen consequences if we do have a lot of devices with TTL=64.
Imagine a "tier-1" carrying some non-trivial fraction of Internet
traffic who is label-switching global table, with no TTL-propagation
into MPLS, and so looks like a single layer-3 hop today. In response
to traceroute-whingeing, they turn on TTL-propagation, and suddenly
look like 10 layer-3 hops.
Having been in the show/hide MPLS hops internal debate at more than
one employer, I'd expect flipping the switch to "show" to generate a
certain support load from people complaining that they are now "more
hops" away from something they care about (although RTT, packet-loss,
throughput remain exactly the same). I wouldn't have expected to
break connectivity for a whole class of devices.
Regards,
Tim.
