So, they use the internet, which is generally considered "two nines" reliable, to achieve 99.99% reliability? Oh, tell me more. <wonka-meme />
I skimmed through the youtube preso and saw nothing about results observed in production, either through actual DC outages, or chaos monkey, or game days.
Seems highly overwrought architecture astronaut'ing and also very difficult to get any telemetry on how well/if it works, to me.
From http://www.nytimes.com/2011/01/09/business/09digi.html?_r=0: "A home user’s Internet connection, with a laptop using Wi-Fi, would be available about 99.8 percent of the time, estimates Mr. Hölzle at Google, which equates to about 18 hours of cumulative downtime a year."
So yeah, two nines.
Also, here's the results of me pinging google.com from the network of my "large technology company" who employees me.
We're talking about mobile internet service. "Two" is probably doable. "Four" OTOH would be less than 53 minutes of downtime a year, which I might believe if you were standing still, on a high point, in an area that didn't have much going on electrically. I would not expect to see that while driving, or perhaps I mean that I have not seen that while driving. But GP is wrong anyway; see sibling comment.
By spreading data to enough mirrors, one can achieve arbitrary reliability. I doubt that any part of their business is only replicated to any particular device. After all, batteries die all the time, unpredictably.
I skimmed through the youtube preso and saw nothing about results observed in production, either through actual DC outages, or chaos monkey, or game days.
Seems highly overwrought architecture astronaut'ing and also very difficult to get any telemetry on how well/if it works, to me.