Web Cache Redirection using a Layer-4 switch: Architecture, issues, tradeoffs, and trends

1. Web Cache Redirection using a Layer-4 switch: Architecture, issues, tradeoffs, and trends Shirish Sathaye Vice-President of Engineering

2. Outline • Web-Cache deployment options - descriptions and tradeoffs • Proxy caching • Transparent proxy caching • Transparent proxy with Web Cache Redirection • Web-Cache-Redirectors - Why do I need yet another device in my network? • Performance • Intelligence • Cache hit-rate • Availability • Summary

3. Proxy caching • Browser pointed at cache instead of origin server • No impact on non-Web traffic • Incremental hardware/software costs limited to cache server and software • Every browser needs to be re-configured • Each client hits only one cache • Can’t take advantage of data stored in other local caches, lowering hit rate • Lower hit rates mean user delays and unnecessary WAN traffic • If cache is down, user loses web access until browser reconfigured

4. Transparent proxy caching • Browser sends requests for web pages to origin server • Impact on non-web traffic: Cache sits in data path, examines all out-bound packets, intercepts and processes web traffic • No browser reconfiguration required • Caches must be configured to do network address translation • Each client hits only one cache • Difficult to take advantage of data stored in other local caches, lowering hit rate • If cache is down, user loses all Internet access until alternate path operational

5. Transparent proxy caching with WCR • Browser sends requests for web pages to origin server • LAN switch (Cache Redirector) in data path examines all packets and redirects web traffic to cache(s) • Very little impact on non-web traffic • No browser re-configuration required • Cache need not do NAT, the redirector can offload this function from the cache. • Each client hits multiple caches • Takes advantage of data stored in all local caches, raising hit rate • Higher hit rates mean less user delay and less unnecessary WAN traffic • If any cache is down, traffic directed to other caches

6. Why do I need yet another device for Web-Cache-Redirection? • Performance: specialized device for high-performance web-cache redirection • Cache-hit rate: specialized device implements techniques to maximize hit rate • High availability: support for redundant caches and redundant redirectors • Intelligence: support for content-aware, content-specific, and site-specific caching. Support for various data types and protocols

7. Maximizing Performance • Per-session: • Every session must be mapped to a cache server • Device must maintain state information in each direction • Must consider special application requirements: eg FTP caching, etc. • Per-packet: • Must substitute session address (NAT) • May need to modify content (header fix-up) • Needs to recompute protocol check-sums • Background: • Must perform server and service health check • Must track load on a per-cache server basis

8. Redirector - Performance Requirements • Scalable performance across all ports • Ability to look arbitrarily deep into packet and rewrite portions of packet content at wire-speed • Ability to support a rich set of redirection rules • Ability to have line-rate performance with richest combination of redirection rules enabled Traffic not subject to redirection experiences minimal latency or throughput impact

9. RISC RISC HW assist HW assist HW assist RISC RISC RISC RISC RISC RISC 1MB 1MB Redirector Implementation Management Module Multi-Gigabit switch backplane 2MB Flash 4MB SRAM • Hardware acceleration at each port for parsing packets at line-rate and performing sophisticated transformations • Distributed processing at each port for flexibility to add new redirection rules with no performance impact • ASIC integrates two 100 MIPS RISC processors, and 10/100/1000 Ethernet MAC per port • Support for traditional L2 and L3 switching at wire-speed • Separate processors for background management functions ... Switch Modules 1MB ports ports ports

10. Intelligence • Network address translation: • Offloads NAT from web-cache • Authentication: • Web-sites may use client source-IP address based authentication • Redirector can be dynamically programmed to not redirect connections for non-cacheable sites • Content-driven caching: • Some information is non-cacheable (eg: POSTs etc) • Redirector can be programmed to not send this info to cache • Redirector can be programmed to not redirect some content types • Support for caching different application protocols • FTP, NNTP, Streaming Audio, Video

11. Maximizing Cache Hit-rate • Variety of content distribution algorithms for a cache array • Balances load across caches using load-balancing techniques • Controls degree and placement of replicated information depending on which cache selection algorithm is used. • Allows tradeoff between hit-rate, performance, replication and fault-tolerance. • Examples of cache selection algorithms: • Hashing on origin server IP address • Transforming the origin-server IP address and some portion of client address using a deterministic function • Using a least-loaded-first scheme • Using a round-robin scheme

12. High availability • Redirector monitors cache health • Supports dynamic cache addition and removal from array • Automatically redistributes content between remaining caches • Supports redundant switching topologies with no single point of failure

13. Summary • Transparent proxy with web-cache redirection is often the preferred way to deploy a web-cache system • A specialized Layer-4 switch (Redirector) offers significant advantages compared to redirecting using traditional networking devices: • Performance • Intelligence • Cache-hit rate • Availability