Wednesday, September 14, 2005
I just noticed something that at first seemed very odd, but as I think on it, it does make sense, but it's not intuitive at first (at least to me).
I did three traceroutes from Casa New Jersey. The first to my workstation (in one network block), the second to the office firewall/router (in a second network block) and the third to the company router (in yet a different network block). I would have expected all three to take a similar path through the Internet, but the practice was different than theory. In the diagram below, the top circle represents my computer at home, and the blue path goes to my workstation, the black path to the firewall and the red path to the company router:
Why the (in one case, wildly) different routes?
Because each destination is in a different block of IP address, each of which is handled differently in different carriers due to different peering arrangements—the office block is owned by one of our providers, the firewall block is owned by Smirk (he managed to snag a Class-C network block a long time ago) and the company firewall block is owned by another of our providers.
But that still doesn't explain this though:
The blue one is again from my home computer to my workstation (and yes,
there are two additional hops in this one—perhaps some routing changed
between the time I made the two images) but the other one, the red one, is
to another machine here at the office—the IP address of my workstation is
XXX.YYY.224.11 and the
other machine is
XXX.YYY.227.9, so while technically they're in
different Class-C networks (a “/24”) they're routed to us as a
“superblock” (a technical networking term) of /20. But as to why
the wildly diverging paths, I don't know.
We control IP addresses
XXX.YYY.239.255 and as I mentioned, not all the addresses in that range
follow the same path from my home computer to the office. I was curious so
I decided to map the paths:
This image shows the overall paths, with the double circles the termination points of the various IPs and as far as I can tell, it looks like the “/20” we have is being routed as 16 individual “/24” networks (aka “Class-C”) through out different providers (some through one, some through the other).
Curiouser and curiouser …
Update on Thursday, September 15th, 2005
Smirk emailed a response to this post:
From: Smirk <XXXXXXXXXXXXXXXX>
To: Sean Conner <email@example.com>
Subject: bandwidth mystery
Date: Thu, 15 Sep 2005 03:35:54 EDT
I think if you check with Dan, we are routing various IP's with preferences for different carriers in order to have our actual bandwidth usage line up with our financial commitments to our carriers.
Perhaps that may solve the mystery?