A High-Availability High-Performance E-Mail Cluster

摘要

Rice University has recently completed deployment of a new campus mail system. This new system was designed with sufficient redundancy to survive multiple hardware, networking, and power failures. It was also designed to provide a level of performance that should keep us ahead of the demand curve for several years to come, while providing additional services such as virus scanning and spam filtering. In order to make use of some new technologies, we had to develop new software to route POP and IMAP connections. This paper is an analysis of our previous campus mail servers, why they failed, and a discussion of the architecture of the new system. Like most university environments, we have seen the explosive growth of e-mail services, with little or no growth in budget to provide these services. The Rice UNIX Facility, or RUF, which handles faculty and staff e-mail for approximately 5000 accounts, has no capital budget allocation. All hardware for this system was donated by the educational network. Our original mail server, a Sun Ultra 1, was adequate for the demands of the campus through 1997. In 1997, a four processor SPARCcenter 1000 became available and was immediately pressed into service. At this time, we noticed a shift from filesystem mail readers like Elm and Pine, to remote mail readers using POP, and to a much lesser degree, IMAP. The loads on that system were largely NFS and mailbox sizes were small. That system kept pace with demand until early 1999, when we were forced to a newer machine due to lack of disk space.