Network Computingwww.networkcomputing.com

Yet despite the merits of ICS and DynaCache, they don't fill the tall order of providing a completely reliable and fault-tolerant configuration that is invisible to the end user. A failure in a ICS cluster will interrupt traffic for the brief duration of the occurrence, and the InfoLibria solution is also susceptible to service disruptions.

Print this Page E-mail this URL

Layer 4 switches can handle complex configurations, but they can be expensive--$10,000 to more than $50,000 for redundant setups. If you can afford to go this route, you'll gain the ability to add and remove cache servers as well as control the amount of cache traffic each handles. Web traffic passes through the Layer 4 switch and is intelligently routed to the most available Web cache server. Cache availability is computed using numerous weights and measures ranging from response time, ICMP (Internet Control Message Protocol) ping and cache server-side agents. When a cache fails, the switch simply stops sending additional Web content requests to that specific node until it is detected as online again. Layer 4 switches also include management options that allow routine system maintenance without disruption of service to Web clients.

Networks equipped with Cisco products have the alternative of using WCCP as part of a caching solution. Like a transparent cache, all Web traffic that passes through the Cisco router with the WCCP feature enabled is redirected to a cache that supports WCCP. When a cache fails, WCCP does not send any additional requests to that cache until it is back on line. Currently, caches that support WCCP are available from Cisco, Novell, Inktomi Corp., InfoLibria and CacheFlow.

Creating a fault-tolerance caching solution generates a new problem. With content requests being distributed across multiple caches, there's no longer a guarantee that Web content requests will be directed to the cache with the desired Web object. Layer 4 switches handle this situation by remembering which cache served the object request for a specific URL or Web site. Much like load-balancing Web sessions, the Layer 4 switch uses that memory to treat subsequent requests. Other setups such as WCCP employ a hashing algorithm to negotiate the proper cache for a request.

As for a cluster of cache servers without an intelligent device, such as a Layer 4 switch, ICP (Internet Caching Protocol) is used to communicate requests between multiple servers. Although it is very simple in nature, it is an effective approach to employing all the cache cluster's resources.

When a cache receives a request for a Web object that it does not have, it sends a multicast request to the other ICP-enabled servers with which it has peering relationships. If the requesting cache does not receive a response within a brief period, it begins to retrieve the Web content from the origination host. If any neighbor caches respond with a hit message, the Web object is then retrieved from the cache with the fastest hit response. Likewise, if all the neighboring caches respond with a miss message, the cache then retrieves the content from the originating server.

ICP is ideally suited for a hierarchical cache configuration with siblings and parents. When satisfying requests, an ICP cache will first ask its siblings for hits and misses. If no sibling has a hit and the cache has a parent cache, it will ask the parent cache to retrieve the content. If the parent has a cached object, the Web client request is immediately satisfied. If it does not, the parent cache can be asked to retrieve the content for the requesting cache and deliver it to the Web client.

ICP is extremely easy to set up and integrate in a network environment. Multicast is supported, and ICP's ability to work with multiple caches in multiple configurations can accommodate virtually all scenarios.

Send your comments on this article to Gregory Yerxa at gyerxa@nwc.com.