in Futurist, Raleigh, Reviews, X-Geek

Google to Google routing could be better

fiber_houseLike many Triangle residents, I’ve been eagerly awaiting Google Fiber service, ready to ditch my indentured servitude to Time Warner Cable. I’m a fairly advanced geek, too, hosting this site and others on Amazon Web Services. I want my website to be as speedy as possible to me and my web visitors, so low network latency is very important. For those who aren’t advanced geeks, network latency is how long it takes for a packet to travel between two points on a network, usually measured in milliseconds. Networking often hits upon the limitation of the speed of light (or radio propagation, depending on the medium), meaning a server located far away (like Singapore) will have a noticeable delay for visitors in America.

My Amazon virtual server is physically located in Ashburn, Virginia but due to some favorable network routing it responds very quickly in the Triangle area, almost as if it were right across town. I have found it very hard to find a server that’s any closer, network-wise.

Now, I’ve been generally happy with Amazon’s service but I don’t like having to pre-pay for a virtual server in order to get the price down to an affordable rate. I also like competition, so I’ve tried kicking the tires of Google’s own cloud service, Google Compute Engine. Compared to the latency of Amazon’s service, though, Google Compute Engine had twice the network latency (or more!) from my Time Warner Cable cable modem at home. Far too many network “hops” stand between me and my Google server.

Enter Google Fiber. I had hoped that a Google Fiber connection would provide a blazing-fast link to Google’s Compute Engine cloud, since Google owns both and shouldn’t have to ride any other provider’s network to get there. I was disappointed this morning to see this is not the case. Checking out a traceroute posted yesterday by one of the Triangle’s first Google Fiber customers, I picked out their external IP and performed a traceroute from a Google Compute Engine instance I spun up today.
Here’s the traceroute posted by Google Fiber customer “undyingfire” from his home to

traceroute to (, 64 hops max, 72 byte packets
1 ( 0.373 ms 0.179 ms 0.137 ms
2 ( 0.330 ms 0.461 ms 0.957 ms
3 ( 2.458 ms 2.079 ms 2.479 ms
4 ( 2.724 ms 1.756 ms 2.471 ms
5 ( 2.903 ms 2.349 ms 2.788 ms
6 ( 83.741 ms 88.358 ms 83.624 ms
7 ( 11.973 ms 11.521 ms 12.298 ms
8 ( 24.266 ms 23.730 ms 24.213 ms

Picking out the first publicly-available IP, I tried pinging it from my GCE instance:

[markt@instance-1 ~]$ ping
PING ( 56(84) bytes of data.
64 bytes from icmp_seq=1 ttl=55 time=33.6 ms
64 bytes from icmp_seq=2 ttl=55 time=33.2 ms
64 bytes from icmp_seq=3 ttl=55 time=33.2 ms
64 bytes from icmp_seq=4 ttl=55 time=33.1 ms

Thirty-three milliseconds. Not terrible, but not blazing fast, either. Still 50% slower than my Amazon route to Ashburn.

The traceroute shows a modest 5 routers between GCE and the Triangle Google Fiber:

[markt@instance-1 ~]$ traceroute
traceroute to (, 30 hops max, 60 byte packets
1 ( 12.601 ms 12.584 ms 12.548 ms
2 * * *
3 ( 77.204 ms 77.352 ms 77.196 ms
4 ( 77.672 ms 77.875 ms 77.781 ms
5 ( 45.213 ms 45.148 ms 45.084 ms

Forty-five milliseconds to this router? That’s pretty slow. I would expect this if I were pinging a server in Seattle but this host is supposedly on the east coast. What gives?

This would be the killer application for Google, blazing fast connectivity between their Google Fiber users and their own Google cloud, yet it leaves much to be desired at this point. Amazon isn’t in the broadband business so if Google wanted it could put a significant dent in AWS’s market share simply through better peering. It seems they haven’t planned or implemented this at this point.

Fortunately, Google Fiber “makes it up in volume” with its huge capacity for bandwidth. It would be nice, though, to see lightning-fast latency between the two Google products. Hopefully Google will make this happen.