Network Computingwww.networkcomputing.com

Print this Page E-mail this URL

Transparent Caches
Transparent caches pose an elegant alternative to proxy-based caches, but there's a significant downside. In a transparent configuration, a caching appliance intercepts all outbound network traffic including HTTP. As long as all HTTP traffic passes through the Web cache, clients do not need to be specifically configured to direct Web traffic to the cache.

There are three standard ways to place these caches inline: as an Ethernet bridge; as the network's configured default gateway; and by using a protocol such as the Cisco Systems-authored WCCP (Web Caching Control Protocol) to transparently intercept traffic at a router or switch. Your network's configuration usually dictates which method is most well-suited your existing network. Adding a cache as a bridge or new default gateway also introduces a new single point of failure for all traffic in addition to Web traffic, so administrators may be inclined to consider a proxy-based alternative. Using a WCCP-enabled switch or router avoids introducing a single point of failure but may also increase load on an integral piece of network hardware.

WCCP also employs a cluster-like configuration of Web caches. When an incoming request is intercepted, it is redirected to a specific Web cache based on the Web URL. This can lead to some caches serving most of the content if Web traffic is limited to a small number of sites.

An obvious benefit of a transparent cache is that you need not configure Web clients to explicitly use Web caches, which makes this option preferable for large ISPs and enterprise environments. The downside is that each Web client is unable to avoid a failed Web cache or even detect a failure, and other network traffic is equally affected by a failure.

Fault Tolerance
In the event of a cache server failure, you can't reroute traffic around it if the cache is transparent. Thus, fault tolerance must be introduced to detect failure and direct the Web traffic. The same holds true for proxy-based solutions.

Some vendors already incorporate failover mechanisms, such as clustering options and solid-state failover devices. In our recent tests of caching servers, we evaluated Novell's Internet Caching System (ICS) and InfoLibria's DynaCache (see "Speedy Performance, Rock-Bottom Price Put Squid Freeware on Top" at www.networkcomputing.com/1011/1011r2.html). Both solutions provide their own failover mechanisms.

The ICS appliance can be configured to cluster when more than one ICS is available. While this solution is capable of circumventing some failures, it is unclear whether all Web requests eventually will be served during a failure. ICS clustering works in conjunction with round-robin DNS: While client proxy requests are shuffled between each ICS in the DNS record, each ICS appliance periodically lets the rest of the cluster know it's still working properly. When a ICS in a cluster fails to send an update, another ICS within the cluster assumes its identity.