Information Services and Technology is considering revamping the MIT e-mail system to include calendaring. IS&T is also considering changing the mail system’s infrastructure to include commercial products like Microsoft’s Exchange server, even while expressing serious concern that those products may not scale to function adequately in MIT’s demanding e-mail environment. Currently, the mail system is based on open source software, though it includes commercial devices for spam filtering.
With seven months past since MIT’s last large-scale mail system failure, there continue to be concerns that those failures could happen again. IS&T is planning to deploy more storage hardware and mail servers to alleviate the problems that led to filesystem recovery taking more than a day to complete — but those changes have not yet happened.
Concurrently, IS&T continues to look at the evolution of spam-filtering software and is piloting an evaluation of Symantec’s Brightmail spam appliance with the Sloan School. That system could eventually replace the Barracuda spam appliances that currently filter MIT’s e-mail.
Post office and calendaring
According to Jeffrey I. Schiller ’79, Network Manager, IS&T has formed a team to look at “issues and options” related to e-mail and calendaring services. Schiller expressed dissatisfaction with the TechTime calendaring service — IS&T’s recommended calendar software — which does not have tight integration with e-mail clients.
The servers in MIT’s mail system that store received e-mail and are directly accessed by users are called post office servers. The post office servers currently use open-source software called Cyrus, developed at Carnegie Mellon University.
Schiller said that several of the software products being looked at might support both post office and calendaring functionality.
The spreadsheet of options IS&T has been considering includes the open-source software Cyrus and Zimbra as well as commercial software from Microsoft (Exchange 2007), Scalix, and Oracle. IS&T is also considering outsourcing the management of mail systems to external parties like Google and Microsoft.
Schiller said that “we haven’t chosen a technical direction at this point” and that IS&T would be “approaching members of the community to get their input.”
Schiller expressed worries about running a mail system infrastructure based on Microsoft’s Exchange server. “To build something [to the] scale of MIT, how many Exchange servers do we have to have? Are we really going to have three people [dedicated to supporting the mail system]?”
March failure could happen again
In March 2007, the post office server po14.mit.edu suffered a catastrophic failure that left thousands of users without access to their e-mail for several days. IS&T diagnosed the problem as architectural: the file storage systems associated with the post office servers were too large. In the event of a failure, the filesystem recovery software took over a day to run, without any evidence that it would ever complete.
Mark V. Silis ’99, who manages network services for IS&T, said that the failure from March could very well happen again. Silis said that IS&T had been planning the evolution of the mail system for some time and had taken delivery of a new storage server. IS&T had hoped to transition to it “in the coming month or so,” Silis said.
According to Schiller, the new storage server would be configured with 500 gigabyte filesystems, rather than the current 1,000 gigabyte filesystems. With the smaller filesystems, Schiller said, he expects the filesystem recovery software to take 90 minutes in the event of a failure, rather than the longer-than-one-day recovery that was experienced in March.
Schiller also said that by automatically purging users spam e-mail from the mail servers every 21 days, IS&T had been able to reduce the number of files in each filesystem, which would hopefully have a dramatic effect on filesystem verification time. Schiller said that though he wouldn’t want to run the filesystem recovery tool, “it probably would terminate in finite time.”
Schiller said that no tests had been done to measure the speed of filesystem recovery but that it would be possible to do such tests.
MIT’s spam filtering has been based on a spam-filtering “appliance” from Barracuda Networks since August 2006. During that time, Schiller said, there have been periodic problems with the Barracuda devices, generally requiring manual updates to their firmware in response to a new style or kind of spam activity. One of the benefits of the Barracuda devices was supposed to be their automatic updates to respond to changes in spam.
IS&T has been testing another spam filtering technology: Symantec Brightmail Anti-Spam. Silis said that IS&T had partnered with the Sloan School to test the Brightmail system on the sloan.mit.edu e-mail system, and that while “there is no definitive plan right now,” they would evaluate future directions for campus spam filtering at the end of the calendar year.
Schiller noted that one of the advantages of the Barracuda system is that pricing is per-appliance. The Brightmail system is priced per-user, and could well be cost-prohibitive for the entire MIT e-mail system. “We are in negotiation” over pricing, Schiller said.