Timwi wrote:
As already mentioned, the latency is negligible (and indeed between
Europe and North America there is often no more latency than
intra-continentally). As for catastrophe preemption, I figured if
LiveJournal -- a commercial business -- isn't doing it, it can't be
*that* much of a risk. Of course I remember those outages, including
LiveJournal's, but we never really lost any substantial amount of
data. The hurricane wouldn't have caused any more trouble than us
having to rewind to the last backup (which a lot of people world-wide
had downloaded, as far as I understood it).
Timwi
Western Europe is relatively small with a high population density, and
has *very* good ping times, typically order of 20 to 30 ms, including
things like ADSL interleave. Here are some actual measured figures for
comparison:
From London to the Paris cluster:
64 bytes from
chloe.wikimedia.org (212.85.150.132): icmp_seq=3 ttl=53
time=28.4 ms (consumer ADSL)
64 bytes from
ennael.wikimedia.org (212.85.150.131): icmp_seq=3
ttl=54time=10.8 ms (Ethernet connection at ISP PoP)
Compare with London to Florida ping times:
64 bytes from
www05.wikimedia.org (207.142.131.247): icmp_seq=1
ttl=49time=121 ms (ADSL)
64 bytes from
www09.wikimedia.org (207.142.131.204): icmp_seq=5
ttl=54time=105 ms (Ethernet)
Now, if a connection was being limited by TCP startup alone, this means
that data would be delivered somewhere between four and ten times
faster, and in particular that page-rendering would be able to start
four to ten times as quickly. In practice, this is not the sole factor
involved, so observed gains will not be as great as this, but they will
certainly be very noticeable by users.
-- Neil