Sometimes, you get a little piece of information that is exciting for no reason what so ever, this is one of those times.
On December 3rd, Amazon Web Services announced that US-West zone was officially open for business. This was very exciting to me, because it is close, and I believe that the West coast is better than that “other” coast. They went full tilt with it, having basically every AWS cloud service available on launch.
It was always a small bit of amusement to me that AWS was centered on the east coast with the US-East zone in North Virgina, when Amazon.com itself is based in Seattle. After US-East was the EU-West zone in Ireland, which made sense since it is a major popular hub (for Europe) and it gives great geographic diversity. Finally, though, we have US-West here in good old Northern California, or more specifically in the Bay Area.
The first thing I did was warm up an EC2 instance (as it happens, I had other testing to do, so serendipitous timing) and ping it. I got ping times as low as 8ms, but averaging around 11ms. The next thing I did was ping (from US-West EC2 instance) one of my favorite places to ping test: Hurricane Electric. They’ve got 2 locations in Fremont and 1 in San Jose.
The average ping time between EC2 and HE.net?
1.6 ms.
From those ping times, I’d suspect that either the US-West zone is IN the Hurricane Electric facilities, or RIGHT next door. It would make sense since they’ve got two availability zones (one in Fremont and one in San Jose, maybe?) which could use HE’s existing facility interconnects for free. I’ve also included the traceroute:
traceroute to he.net (216.218.186.2), 30 hops max, 60 byte packets 1 ip-10-160-20-2.us-west-1.compute.internal (10.160.20.2) 0.652 ms 0.640 ms 0.628 ms 2 216.182.236.73 (216.182.236.73) 51.287 ms 51.386 ms 51.551 ms 3 72.21.222.228 (72.21.222.228) 1.290 ms 1.398 ms 1.510 ms 4 72.21.222.195 (72.21.222.195) 2.008 ms 2.150 ms 2.145 ms 5 paix.he.net (198.32.176.20) 1.840 ms 1.971 ms 2.111 ms 6 10gigabitethernet1-2.core1.fmt1.he.net (66.160.158.241) 2.099 ms 1.870 ms 2.102 ms
Since it was worth investigating, I dug up more big colos in the Bay Area, starting with ColoServ, located in San Francisco itself, which had an average ping time of about 5ms. I found another in San Jose, namely Silicon Valley Web Hosting with which I got a ping time of about 3ms from EC2. I started to run out of local colos that I knew an address to ping, but I did give ServePath a shot too. They came in just under 4ms average.
Seeing as how a ping time of 1.6ms to he.net is astronomically low, a traceroute that shows packets basically never hitting the public internet, with the exception of going through PAIX… my money is on HE.net. More specifically their Warm Springs (Fremont 2) location, since it is 200,000 square feet. Supposedly it is (or was when it was built) the largest colo facility in the Bay Area. I could be wrong, but I know for a fact that wherever US-West EC2 is, it is in the Fremont to San Jose area. Ping times for anything else are just too high. Plus, why would Amazon pay more for hosting in San Francisco if they didn’t need to?