Studies of companies of different sizes showed that the relationship between size and cost of capacity could be approximated into a diagram which would serve as a starting point for discussions around scale advantages. Among others Ronny Wester, then at Volvo Data, wrote a few articles on the topic.

Conclusions for Mainframe Operations

The diagram shows that as the volume gets bigger the marginal cost improvement gets smaller. However, large volumes and large data centers are more difficult to manage and operate. The differences in cost allocation between data centers in the study showed that for large data centers the ability to manage and control operation was more important than the actual scale advantage.

At the end of the 1990s and the beginning of the 2000s we had the opportunity to study the mainframe operation in greater detail. We found that there was a big difference in utilization between data centers. The commonly accepted scale advantage was only applicable on nominal size. A study of cost per used capacity gave us an altogether different picture. Even small data centers with high utilization could compete with large ones. When we furthered the study to look at cost per capacity utilized for customer use the importance of data center management became even more obvious. Data centers with quite small nominal capacity could prove to be up to 25% cheaper than large data centers when looking at cost for capacity used by customers. The key was to measure load from applications used by customers and exclude the data centers own use and tools.

Another obvious reflection is the ability of the data center to recruit and retain a good management team. In a small data center the demands on management are not as high and even with limited experience and ability the data center can be managed without too much impact on efficiency. Simply put - a large data center can be managed a lot better, or worse, than a small.

What conclusions can be applied to a modern server farm?

Let us first look at what a modern data center is. Before the virtualization of server farms large scale operations in the tradional sense existed only to a limited extent. Large server farms consisted mainly of groups of servers dedicated to different customers where the customer often owned the server or paid the actual cost of the servers used.

With the introduction of virtualization the situation changes dramatically. Server farms becomes more and more like mainframe data centers. Virtualized server farms are in terms of ownership, management and control managed the same way as mainframes. The continuos development of virtualization as we see it today will just reinforce the comparison. A discussion around scale advantages is therefor relevant also for server farms.

The above discussion on mainframe has the same conclusive roll for the server farm as for the mainframe data center. Management, control and capacity utilization is just as important for achieving a low cost per produced service. A smaller virtualized server farm can be as competitive against a larger farm as a smaller mainframe data center to a larger.

Taking into account the utilization, traditional measures in server farms, like "the number of servers per system administrator", are irrelevant to describe the cost of IT operations and the service delivered.

Managing a large server farm might even require more from management than a mature mainframe data center which further reinforces the argument of managements importance to achieve scale advantages.