A balanced computer system needs about 1 MB of main memory and 1 megabit/sec of I/O bandwidth per MIPS of CPU performance.
A program executes about 90% of its instructions in 10% of its code.
Bandwidth grows by at least the square of the improvement in latency.
The miss rate of a direct-mapped cache of size N is roughly the same as a two-way set-associative cache of size N/2.
Design with no single point of failure.
In 2011, the fully burdened cost of 1 Watt per year in a Warehouse Scale Computer in North America—including amortized power and cooling—was about $2.
