Automatic and Reliable Elimination of E-mail Loops Based on Statistical Analysis

American Medical Informatics Association 1996 Annual Symposium

医药卫生;医药卫生

E. Solana ; V. Baggiolini ; M. Ramluckun ; J. Harms

Others  :  http://
PID  :  47167

来源: CEUR

【 摘 要 】

One of the major threats to the E-mail service are message loops, mail messages which endlessly circulate between a number of E-mail relays and severely deteriorate or even interrupt the messaging service. Although the standards ruling E-mail communications include mechanisms to prevent message loops, these mechanisms are often unimplemented and, if they are, can fail to work reliably under certain circumstances. This paper describes a novel approach to the automatic and reliable elimination of looping messages. Statistical properties of the message traffic are analyzed to detect messages that might be looping. Such messages are earmarked, and if they are observed to actually be looping, they are eliminated. This method has been implemented and successfully applied in the multi-protocol E-mail environment of the University of Geneva. Problem description A message is said to be looping if it is forwarded endlessly in a circular manner among a set of mail relays (Message Transfer Agents or MTAs). Normally, two or three mail relays take part in a message loop, but there can be more than that. The period of a loop (the time it takes to complete) ranges from a few sec- onds to a few minutes. Looping messages usually grow in size, since mail relays typically add header lines to messages they process. A message that loops during a weekend can be forwarded tens of thousands of times, growing in size to megabytes and producing megabytes of logging data. The situation is even worse if distribution lists are involved (c.f. [1]): the number of messages explodes and soon the mail system crashes under the load. These scenarios illustrate how important it is to detect and eliminate looping messages automatically. Loops are caused by wrong addressing configu- ration on one or more systems: • bad configuration of mail relays, e.g., a relay that does not ‘‘recognize’’ messages addressed to it and, instead of keeping them for local delivery, forwards them to a central mail relay which sends them back again1. • automatic forward mechanisms such as aliases, distribution lists, or personal forward mecha- nisms (.forward files) pointing to invalid addresses. • automatic reply mechanisms, such as vacation programs or non-delivery reports. The mechanisms that are directly under control of a proficient administrator account for only a small