Server sprawl: Problem or solution?

Larry Seltzer examines whether server consolidation can save you money in terms of power, management, and even the cost of physical space.

Some old ideas are making their way back into technology hipdom. You may have seen IBM's TV ad, "The Heist," pushing their zSeries mainframes running Linux. It's a great ad, much better than those dreadful basketball ads. The point the ad makes is that consolidating many servers into one can save you "a bundle."

This ad is part of a general campaign to promote Linux, but I'm more interested in the idea of consolidation than the actual operating system being used. Server sprawl can be a legitimate concern for a number of reasons. You can make a cost argument for a smaller number of more powerful servers independent of the hardware cost: you might save money in terms of power, service agreements, and even the cost of physical space.

There is a related question of performance growth in your servers. Does it make better sense to have a larger number of cheap, commodity servers, or a smaller number of higher-end, more powerful servers? In the most abstract sense, there are great arguments on both sides and many companies, including both IBM and Microsoft, sell solutions in both spaces. The real answer is that it depends on your situation and your applications, and only someone familiar with your particulars can make an intelligent decision for you.

How do we end up getting server sprawl? Companies will buy and organize servers on an organizational basis: there will be marketing servers, engineering servers, accounting servers, and so on. Plus these servers have gotten cheap. Finally, if you want to be sophisticated about it, you can combine many low-end servers into a cluster that solves large problems and can perform more work in parallel. It's not uncommon, for example, to cluster several Intel-architecture or Sun servers together running a single database application, usually partitioning access to the physical database. A query for a record beginning with the letter "A" may normally go to one system while a query beginning with "S" goes to another.

There's a lot to say for having a large number of smaller servers, my favorite argument being that it gives you redundancy. If you plan correctly, should there be some problem with one of the servers, you should be able to find another with some free capacity and transfer. The hardware is pure commodity stuff, and you can switch vendors at will, for the most part.

But then one day you wake up and you've got so many servers that some vice president starts complaining about your air conditioning and electric bills. And the high-end servers are also coming down in price, and the quality of the components, so I'm told, tends to be higher in them, so you're less likely to have to deal with hardware failures (good thing, since less redundancy means you'll be more reliant on each server).

I can see that there may be cost savings in this regard if you were able to take scores of servers and replace them with a few, but I'm not as clear on how there are any large "management" savings. Unless, at the same time, you completely reorganize your IT infrastructure, you're still going to have all the same management tasks to perform. It's not likely that simply having fewer server systems will mean you need fewer administrators, because most of the same administrative work will remain. And not all administrators have all the same skills; database administration and file/print administration are very different, for example.

Backup is one task that, in most ways, is facilitated by consolidation. It's just easier to backup a smaller number of servers. Of course, these servers will have relatively large storage sets, so you will likely need to acquire new backup hardware as part of your consolidation scheme.

The IBM consolidation ads could use a little more detail in their explanation, by the way. Even though the servers are running Linux (actually, they run multiple virtual instances of Linux), they are essentially mainframes, and a tough sell into any IT shop that has no existing mainframe expertise. In large corporations where mainframes are common and core to the mission, one of these zSeries fridges can be a great way to take projects out of a "start-up" mode and put them into a more mature, reliable environment. Don't get the idea that a standard Microsoft shop would be well-served by one of these. Some of the same stuff is going on in WindowsLand though, and for all the same reasons. It could easily pay to consolidate lower-end database servers from different departments onto one big monster box that still had room to grow.

It strikes me as odd that consolidation should gain currency at a time of relatively low prices. Is server consolidation appealing for your own company? Let me know in the Talkback below. Next week I'll be looking more closely at the performance angle of this same problem, what Microsoft calls "scaling out vs. scaling up."

How does your company approach server consolidation? E-mail Larry or post your thoughts in our Talkback forum below.