Their baseline values are very misleading.
But their relative metrics can still be useful.
For example, it's very inaccurate to say that 51% of the internet is down.
But it's precise to say that packet loss among the working nodes has increased about 30% in the last 24 hours, and sharply.
But that's likely 2 or 3 nodes, not a meaningful sample
Re:App Engine down
by obituary_latte on Friday October 26, 2012 @11:45AM
(#4702680)
Agreed. I was surprised to see only one[1] of the "down" routers with any packet loss over the last 24 hours.
That said, it's still interesting how the overall traffic trends so sharply downwards. I wonder if they have more data than they are showing in the graphs.
Their sample size is extremely small, and most of those are permanently down.
Have a look through their list of north american routers and find one of them where packet loss has gotten worse as their main overall graph for packet loss would suggest - I've just been through them all and couldn't find one.