Footprint Descriptors: Theory and Practice of Cache Provisioning in a Global CDN

https://www.akamai.com/content/dam/site/en/documents/research-paper/footprint-descriptors-theory-and-practice-of-cache-provisioning-in-a-global-cdn.pdf

Abstract

• Modern CDNs: highly diverse set of traffic classes, each with different object access patterns, object size distributions, and cache resource requirements

  • I.e. web content from a specific new site, or image content from a specific e-tailor, or iOS downloads from Apple, video content from Hulu, etc.

• Sharing of cache resources across multiple traffic classes poses significant cache provisioning challenges

• Introduce footprint descriptor, which is a succinct representation of the cache requirements of a request sequence

Intro

• CDN: cache provisioning and request routing at the granularity of traffic class

  • Key decision: which subset of the servers must serve which traffic classes

  • Classes --> different caching characteristics and performance requirements

    • Web content from e-tailer may require fast response time and high cache hit rates, and the object sizes are smaller

    • Background software downloads: larger

• Cache provisioning: determine which traffic classes are hosted in which cache servers of the CDN, given a vast platform of cache servers with varying amounts of cache space available at each server

  • Goal: model the caching requirements of traffic classes and predict the best way to assign traffic classes to cache servers, so as to optimize the use of cache resources and provide an acceptable hit rate at a reasonable cost

  • Input: sizes of the caches available in servers across the CDN, and the characteristics of the request sequences for each traffic class

  • Output: the set of servers that serve each traffic class

  • Offline planning step, online used as a mapping system to route the requests of each traffic class to one of the provisioned cache servers in real-time