Datacomp: Locally Independent Adaptive Compression for Real-World Systems

摘要: Non-lossy compression can save time and energy during communication if the cost to compress send input is less than of sending it uncompressed. Unfortunately, also degrade performance, no single method always beneficial, outcomes depend on many factors. As a result, choices in real systems are coarsely grained manually controlled, resulting suboptimal or even poor performance. Adaptive Compression (AC) make dynamically optimize utility. Existing AC limited ways that reduce their suitability for general-purpose computers. Datacomp an system operates locally includes significant hard-coded knowledge. Using real-world data, broad range environments Comptool "AC Oracle," we show Datacomp's performance equivalent close ideal at bandwidths between 1-100Mbit/s, when static strategies more costly compression. While struggles perform well 1Gbit/s, understanding why illustrates important challenges suggests solutions.