Well, a new study has good news: there's a 1.5% chance that the V.I.P. you directed your e-mail to never even received your little firebomb.
The bad news is that 1.5% of the people you want your mail to go to aren't getting it — and you may never know why.
Disappearing Into the Ether
Here's how Moors and his group tested the basic structure of the e-mail system we've come to rely upon:
• Set Up the Trial. E-mail accounts were first established with two large Web portals (Yahoo and Hotmail), two large Internet service providers (Optusnet and Bigpond), two universities (Brooklyn's Polytechnic University and Moors' own UNSW) and 12 e-mail service providers who offer free e-mail addresses.
• Send Numerous Test Messages. The group used its own computers to send hundreds of e-mail messages to each of the test accounts and check whether the messages were ever received. Each missive was as short as possible — little more than a date and time, such as "Wed Aug 18 08:14:00" — to reduce the risk that messages would be mishandled due to their content.
• Measure the Awful Results. Every receiving service "lost" at least one e-mail during the test. Of the 1.57% of messages that were never received, one-third (0.5% of the total) vanished silently — with no notice whatsoever to the sending party. Furthermore, one of the free e-mail providers lost an astonishing 10% of the messages sent to it over a one-month period. (In a telephone interview, Moors said he couldn't recall the name of that provider.)
The messages above were "dropped" for purely technical reasons. But you can add to their number the percentage that's waylaid by sloppy content filtering. One recent study, which I described in this space on April 19, 2004, estimated that the top 16 U.S. Internet service providers shunted as "junk" about 19% of the legitimate, permission-based business-to-consumer messages that the ISPs' users should have received.
Errant spam filters are a separate problem. Meanwhile, the idea that more than 1.5% of the messages that a university basically sent to itself didn't survive the Internet's e-mail routing system is shocking. This cries out for a quick, technical solution.
Protecting E-mails and Protecting Yourself
Moors believes there are several things Internet users can do to give e-mail a more predictable outcome, ensuring at least that proper notices get back to senders in all cases of failure:
• Fix Send-and-Forward. The Net's existing store-and-forward system, in which each router is responsible only for moving an e-mail message one step toward its destination, "fundamentally can't guarantee delivery," Moors says. "Nothing protects against server failure" under the current infrastructure, he notes. If an intermediate server crashes or simply becomes overloaded, all of the e-mail messages waiting for routing may be lost.
• Fix the Servers. Among other improvements that could be made to today's widely used e-mail server software, Moors suggests the addition of better storage systems. More robust storage, building upon the experience gained by peer-to-peer file-sharing networks, could help servers recover from crashes and overloads that now cause the permanent loss of messages.
• Fix the Clients. There is much that PC client software, such as the dominant Microsoft Outlook program, could do to monitor and repair the fractured flow of mail, Moors says. One step would be to allow senders to define alternate recipients. "If a message can't be delivered to the primary recipient," Moors explains, "perhaps the message should be sent to an administrator." Client software could also highlight important messages that haven't been responded to (indicating that they may never have arrived). In addition, programs should filter out false alarms — such as the bogus "nondelivery notices" that are commonly used by spammers — that aren't in response to any message that the end user ever sent.
Moors emphasizes that his group's research is in its infancy and needs a much larger test bed of both senders and receivers. In fact, his project's goal is actually to test all forms of network reliability, with Internet e-mail simply happening to be a very large network to test.
He invites all interested Internet users, therefore, to volunteer for a more sophisticated experiment. Participants would agree to let the university group know whether any or all of five test messages were ever received. To sign up as a volunteer, use the group's volunteer page.
Next week, I'll report the responses of some Internet service providers and other experts to Moors' preliminary findings.