E-mail is one of the Internets most mature applications, yet todays climate of tighter corporate regulatory controls and increased attacks from malicious and unsolicited mail sources is forcing administrators to look more closely at the way theyre implementing this vital resource.
Your companys strategy for effectively managing e-mail must solicit participation from individual users, whose capacity for trimming away spam and correctly identifying messages that merit archiving far outstrips any algorithm that a vendor might throw at the problem.
While frequently overlooked, the client-side message-filtering features of enterprise e-mail applications offer companies a great opportunity both to build smarter e-mail management solutions by taking advantage of their workers knowledge and to help users get more out of e-mail by better organizing their mailboxes.
However, while the most prominent mail clients available on Windows, Linux and Mac OS platforms offer message-filtering features, they all possess certain liabilities that have limited their use.
For one thing, a client-side filtering feature must, by definition, offer an interface within the mail client through which users may create, modify and apply filtering rules. However, these rules tend to be stored and executed locally. This presents difficulties for users who access their mailboxes through home or mobile computers, which might not carry the same filters as the users primary workstation, or through handheld devices or Web-mail applications, neither of which are likely to support filtering at all.
Microsoft Corp.s Exchange supports server-side filters that users may define using their Outlook client. In the past, eWEEK Labs has used Outlook to configure server-side filters on an account, which we then accessed using a Linux-based e-mail client, for example.
However, not every rule type that you can define using Outlook is supported for server-side deployment. For instance, rules that flag messages as important must be run client-side.
Sieve, an architecture-neutral mail-filtering language that has been proposed as an Internet Engineering Task Force standard, is also meant to address the issue of server-side support for client-generated filtering scripts. (More information is available at www.cyrusoft.com/sieve.)
A handful of e-mail servers, including Sun Microsystems Inc.s Java System Messaging Server, support Sieve, as do several e-mail clients, including KMail, the default mail client of the KDE Project.
All current e-mail clients share another filtering limitation: The mail-handling rules that these applications create lack intelligence and tend to work only in narrow, well-defined cases. For example, its easy to send all mail from a particular sender or domain to a certain folder, but when different sorts of messages from that sender need to go to separate folders, a string of new, increasingly specific rules must be created to produce the appropriate behavior.
Users whove tried to use mail-client-filtering rules alone to stem the tide of unwanted mail in their in-boxes can attest that when messages carry diverse contents and hail from numerous sources, automatic message-handling schemes quickly break down.
One solution to this mail-filter limitation is widening the scope of Bayesian filtering techniques, which currently filter junk mail based on users determinations of what constitutes unwanted messages. These same Bayesian techniques can be employed to sort other types of mail as well.
Weve spent some time testing POPFile, an open-source application that uses Bayesian techniques to organize mail by learning how users organize it, and weve been impressed so far with its performance. The most unattractive thing about POPFile weve encountered so far is its lack of integration with the e-mail clients with which weve tested it.
We hope to see POPFile or something like it built into mail clients as an enhancement to their mail-filtering systems, in the same way Evolution and KMail can link directly to a locally running SpamAssassin service for mail filtering and training.
Senior Analyst Jason Brooks can be reached at firstname.lastname@example.org.