15 Principles Used to Overcome Network Bottlenecks Number P1 P2 P2a P2b P2c P3 P3a P3b P3c P4 P4a P4b P4c P5 P5a P5b P5c P6 P7 P8 Principle Avoid obvious waste Shift computation in time Precompute Evaluate lazily Share expenses, batch Relax system requirements Trade certainty for time Trade accuracy for time Shift computation in space Leverage off system components Exploit locality Trade memory for speed Exploit existing hardware Add hardware Use memory interleaving and pipelining Use wide word parallelism Combine DRAM and SRAM effectively Create efficient specialized routines Avoid unnecessary generality Don't be tied to reference implementation Used In/Networking Example Zero-copy interfaces Application device channels Copy-on-write Integrated layer processing Stochastic fair queueing Switch load balancing IPv6 fragmentation Locality-driven receiver Processing; Lulea IP lookups Fast TCP checksum Pipelined IP lookups Shared memory switches Maintaining counters UDP checksums Fbufs Upcalls