btw, speaking of orders of magnitude:

big cluster in 1990: 4 nodes
big cluster in 2005: 4096 nodes

big "supercomputer" in 1992: 256 CPUs
big "supercomputer" in 2008: 256K CPUs

so, 10 bits in about 15 years in each case. You can fill in the data
points, and project and look at what people are doing and, guess what,
no sign it will stop. There's a cheat here: that big "supercomputer"
is multicore, but still ...

These aren't Google systems. These are systems that are "tightly
coupled", i.e. run one app and only run as fast as the slowest
computer and the longest network latency will allow. Network latency
is measured in units of hundreds of nanoseconds. At any given time,
there's only one  or two of these being built (that we know about).

ron

Reply via email to