Being considered an expert in your field is nice, but finding that you've been misquoted is less pleasant. Recently, I learned that an article on the Lotus Web site cited me as an expert on the weaknesses and fallibility of Exchange Server.
The article's content was less interesting and factual than you'd expect from a major computer company. In its crusade against Exchange Server, the Notes camp had used comments and facts from a presentation I gave at TechEd '99 in Dallas in May and articles that Jerry Cochran and I wrote for the Exchange Server UPDATE electronic newsletter and Windows NT Magazine, respectively.
The Lotus document is "Reliability: A Domino Focus, an Exchange Afterthought" (http://www.lotus.com/developers/itcentral.nsf/ wdocs/408eb43cc29a927b852567f2006e10cd?opendocument). The article refers to some statistics about Compaq's internal use of Exchange that I cited at TechEd. My slide showed that Compaq had roughly 100,000 mailboxes divided across 400 servers. Lotus interpreted this information as follows:
By doing the math, we see that 100,000 mailboxes divided by 400 servers = 250 users per server! Not only does this increase the complexity of the Exchange environment, the total cost of ownership skyrockets. And Compaq is not the only Exchange customer with such a low number of users per server—at least two other major Exchange customers tell Lotus that they have an average of less than 500 users per server. These are total mailboxes per server—not active users.
On the surface, the analysis is accurate. However, anyone who has ever worked on a corporate deployment of Exchange knows that many factors influence the number of servers deployed. Not all these servers support mailboxes. Some are pure routers, deployed in a central hub site, and others are public folder servers. Compaq deploys other servers on a geographic basis to serve small communities of users. In small communities, deploying the network necessary to link clients to servers in another location isn't cost-effective.
In Compaq's case, the 400 servers result from a merger between two of the world's largest Exchange organizations (Compaq and Digital Equipment). Both companies deployed Exchange Server in 1996 when Exchange Server 4.0 and the hardware then available couldn't support as many mailboxes as they can today. During the TechEd session, I explained these factors. I also said Compaq had too many servers for the number of mailboxes but that the company was reducing the number of servers as the merger progressed. Clearly, these points whizzed over the head of the Lotus observer.
Compaq is making progress toward its goal. In September, Compaq operated 330 servers in 23 sites around the world. The largest site supported 81 servers. A typical month's load illustrates the effectiveness of the deployment. Each month, Exchange transmits more than 1.3TB of messages and delivers more than 97 percent of the messages anywhere in the world within an hour. Compaq handles more than 26 million incoming messages per month.
Anyone can miss a point a speaker makes at a conference session. However, quoting an article out of context to prove a point is unacceptable. Lotus clearly believes that Exchange Server's transaction logging model is a weakness. The article says, "One other weakness in the Exchange data store architecture is its transaction logging model. Exchange administrators must balance the transaction log's need for disk space versus its perceived benefit."
Lotus goes on to make the outrageous suggestion that "Exchange experts have even recommended compromising the transaction logs themselves because of their disk space requirements." That statement implies that experts recommend that you turn on circular logging to keep log files from occupying too much disk space. I know of no Exchange expert who would be so foolish and shortsighted to recommend circular logging on a production server. To the contrary, the firmly established best practice is to put the transaction logs on a different physical volume from the database and to allocate copious space for logs. Even on the largest servers, a 4GB drive is enough to fulfill this requirement many times over.
The article then seeks to associate me with the Lotus position by quoting from my Windows NT Magazine article "Exchange Server Transaction Logs" (February 1999):
Circular logging sounds like the perfect solution to Exchange Server logs' disk-space consumption. The benefits are obvious: a reduced disk-space requirement and no need to monitor transaction log disk space. However, Exchange Server keeps logs so you can recover transactions when you have to restore a database from a backup. If Exchange Server reuses the logs, you can't recover the old transactions. If you use circular logging, you might lose data when you restore an Exchange Server database.
The warning is accurate, but Lotus fails to point out that I had earlier made the point that circular logging is really for systems that support small user communities and that I later recommend that you disable circular logging on any server that creates more than five logs a day (i.e., any corporate system).
The article misinterprets other information, too.
According to Microsoft's own slides, Platinum clustering does not support MAPI clients—like Outlook. From Microsoft's 1999 Fusion conference, presentation SAM 116's speaker notes indicate, "This does not work with MAPI clients, only with Internet clients. Positioning is critical since we do not want to tell customers that this doesn't work with MAPI." Microsoft is ignoring the enterprise needs of its installed base of Outlook users by providing clustering for Internet clients only.
Clearly, the author misheard or misunderstood the speaker's point. Clusters and a front-end/back-end server configuration are different. Clusters provide resilience; front-end/back-end server configurations aid scalability. Exchange 2000 Server (formerly code-named Platinum) supports Messaging API (MAPI) client connections to 2-way (Windows 2000 Advanced Server—Win2K AS) and 4-way (Windows 2000 Datacenter Server—Datacenter) clusters, but the current generation of Outlook clients doesn't support balanced connections to front-end/back-end configurations. In other words, you can have a cluster acting as a back-end server and have MAPI clients connect directly to it, but you can't have MAPI clients connect to a front-end server and be proxied to establish a connection to a back-end server. The next generation of Outlook clients will probably support both connection types. The difference is subtle but important and seems to have escaped the attention of the Lotus writer.
I also take exception to the statement that "Microsoft applied a 'move server' bandage to this problem with Exchange Server 5.5 Service Pack 2 in December 1998, which allows two directories to be merged on a one time basis." This quote refers to the Move Server Wizard, and the bandage analogy suggests that the wizard might be a piece of patchwork code cobbled together to do a quick and dirty job of moving servers around. Although the wizard isn't perfect, it does its job—if you prepare well. Evidence of the wizard's success is how the software has enabled companies like Compaq to merge two large Exchange organizations together reasonably easily.
I could quibble with many other statements in Lotus' document, which is essentially a marketing attempt to blacken Exchange as much as possible. I'm not an apologist for Exchange. I just think it's the leading messaging server available for NT today, which is why I work with and write about the technology. I hate when people quote me without permission and inaccurately, especially when the author takes cheap shots at worthwhile technology. Are you listening, Lotus?
P.S.: I discussed these points with a Lotus representative at the Microsoft Exchange Conference in Atlanta in early October, and the representative agreed to remove the disputed references from the Lotus Web site. However, as of late October, the references were still on the site.
