An Analysis of Microsoft's TCO Comparison

Microsoft's TCO comparison document between Windows NT on Compaq and Solaris on SPARC begins with some summary information that draws the conclusion that the TCO for NT is 37% lower than the TCO for Solaris.

Here we see the first problems with this report: Microsoft is comparing NT on commodity hardware to Solaris on proprietary hardware. This will assuredly mean higher costs for the Solaris system for systems and hardware upgrades. Also, the 37% number is per user, not per server. The per server comparison is only 25% lower.

However, Sun environments have typically been in place longer and have had enterprise capabilities for several years. As a result, although the average number of users is almost the same (925 vs 990) Sun systems in certain cases run in more demanding environments:

The average database size on Sun systems is 47% larger than on Microsoft/Compaq systems.
More Sun users (39%) than Microsoft users (25%) are running transactional applications on their servers.
More Sun users (55%) than Microsoft users (46%) are running Web applications on their systems; Sun systems have a higher average number of concurrent sessions (160 vs. 98) and hits per day (3,462 vs. 1,423).

Although Microsoft claims that the NT systems cost 37% less to own, the Solaris systems handled larger databases and more demanding applications connecting to those databases. They also handled 63% more concurrent connections and 243% more hits per day.

This makes one consider that maybe cost per transaction would be a better measure of the cost of the system than cost of ownership. After all, companies don't buy computers for the sake of owning computers. They expect to get a return on their investment by using them to run their business. The more work the computer does, the better the return on the investment. If the Solaris systems cost 50% more to own but do 50% more work, it's a wash.

Microsoft includes costs of value-added software including databases, development tools, applications and utilities, but they fail to list the actual software installed. Therefore, we cannot compare the features or relative quality between the Solaris software and NT software.

Microsoft includes labor costs for many items. Some of these are specified to the one-hundredth of an hour to show four digits of precision. But they are not consistent with precision. They state that the average NT site had 30 NT servers. In trying to remove rounding errors found while analyzing the numbers, it appeared this number was closer to 29.6. Why did they drop to one digit of precision here when they used four in other places?

Microsoft includes in the comparison the costs of upgrading infrastructure items. These costs are detailed in Table 5. The details of these upgrades are missing so it is impossible to determine the nature or necessity of these improvements. Some of these upgrade costs include server hardware and client and server software even though an earlier section of the report detailed the costs of purchasing new server hardware and software. A proportionately large amount of money was spent by the Solaris owners on cabling and wiring upgrades ($9,400 for Solaris vs. $2,840 for NT) with no explanation of the difference. Also included is a nebulous reengineering costs. No detailed explanations for any of these numbers is given.

In summary, Microsoft has not provided enough information in this report for an IT manager to properly assess what it would cost to own either operating system in his or her own datacenter. Since computers are bought to perform some function, the return on investment must be compared to the cost of ownership in order to determine the real cost to the corporation.

An Analysis of Microsoft's TCO Comparison - Part 1

References