Network Computingwww.networkcomputing.com

Netscape Communications Corp. asserts that 30 percent to 60 percent of all client FTP and HTTP requests to the Internet are redundant requests from users who need information that has already passed through their company's Internet gateway. The idea behind the caching element of proxy servers is to capture the most heavily transferred data at the connection and minimize the traffic that needs to pass over tha t link. This strategy increases perceived network performance and discourages unneeded connections over nondedicated links and usage on dedicated links.

You'll also find that the growing browser and back-end support for proxy servers introduces several failover, load-balancing and replication features that keep your servers' internal structure operating more efficiently. These features make the investment in a proxy architecture much more worthwhile. In Network Computing's lab at Syracuse University, we used Microsoft Corp.'s Proxy Server 1.0 and Netscape Proxy Server 2.5 to test different implementation strategies for both enterprise and small-office implementations.

Cache What? In most environments, the proxy server sits at the Internet link where all Internet requests are forced to pass through it. The caching feature specifically deals only with FTP, gopher and HTTP traffic, and vendors have their own methods for determining what tra ffic to cache. By default, Microsoft's Proxy Server does not cache HTTP traffic that includes a "?" in the URL, which is most often the result of a search. It also omits any secure traffic that was passed to the user with the Secure Sockets Layer (SSL) protocol.

Netscape's Proxy Server, on the other hand, stores SSL authenticated traffic on the proxy server. You must reauthenticate to gain access to the data stored at the proxy server, but Netscape's product saves you from reaccessing the original source.

Spreading the Cache In large organizations, you can incorporate proxy servers into two different, yet common, configurations throughout the enterprise. The first and most obvious position is at the Internet gateway, where traffic incurred by repeat connections can be reduced and bandwidth becomes available for other traffic. One of the more popular placements for proxy servers is on both sides of a leased-line link for private corporate networks that generate significant, static intranet data. Distributed organizations that use slow, leased-line se rvices typically allot only enough bandwidth for high-priority applications.

Additionally, with multiple proxies running in any organization, proxy replication reduces the amount of traffic that must leave the local network. Replication can distribute data to several different proxy servers--whether internal or external--which are closer to the user, reducing local network traffic.

Netscape Proxy Server offers hierarchical replication, which chains together the servers. The data isn't continually replicated from server to server, but when a user makes a request, it is passed on to the proxy servers in the chain to see if the Web page is cached in another location. The requested page is then replicated to the user's primary proxy server. This is useful for branch offices that send Internet and intranet requests through one central point, such as a central office. The branch office can easily mirror the proxy servers running at the central site.

Alth ough Microsoft Proxy Server doesn't offer any serv er replication features at this time, we tested a beta of version 2.0 that offers both hierarchical caching and a unique Microsoft feature, array-based caching. This capability lets the administrator view a cluster of proxy servers as one logical entity with a single virtual cache. In array caching, each server continues making requests to other proxies. The load for page requests can balance over multiple servers because members of the cache array are part of one large cache. This reduces the total amount of disk space required across all proxy servers included in the array, and permits the cache to be managed from one central point.