Virtualization is another way to consolidate. Instead of having a high number of servers using fractions of their available resources, you can virtualize them on fewer more powerful machines and save both on hardware costs and power.
If you follow this line of thinking the next step becomes almost automatic -- if you combine blade, SAN, and virtualization technologies you can create ultra-dense server deployments that multiply the benefits you get from any one solution.
What is ultra-dense server deployment?
The goal of an ultra-dense deployment is to fit as many servers as you can into as little space as possible. Along these lines blade servers are ideal since with some blade models you can fit the equivalent of 16 1U servers into 10U of space.
In many cases the blade servers are just as powerful as their 1U or 2U counterparts, so the extra density you get with blades comes without sacrificing power. Of course, blades typically have less disk expansion options, but we will consolidate the storage on the SAN anyway, so that's less of a concern.
Next, add virtualization to the mix. Virtualization of your servers will allow you to fit anywhere from three to thirty-three (or more) virtual machines on a single physical machine. The amount of additional server density you can get from virtualization will of course vary wildly depending on the horsepower of the blades you have installed and the hardware requirements of your virtual machine. The point remains though, if each of your servers can house even four virtual machines, you have just increased your server density four times.
Ultra-dense server deployment gets help from virtualization software
Ultra-dense servers give even more benefits if you use some of the more advanced virtualization features such as VMware's VMotion combined with Distributed Resource Scheduler (DRS). VMotion allows you to migrate a virtual machine from one physical host to another without any downtime, and DRS takes this a step further and actually load balances virtual machines across multiple hosts for you so you can ensure all of your physical machines share a fair amount of the load.
One requirement of these features, though, is that the physical hosts have very similar hardware. With a blade deployment you already know the servers are identical--that's part of the point--so VMotion and DRS just work without any tweaks. Plus, expansion is easy--slot in another blade, add it to the cluster occupied by the rest of the blades, and DRS will take care of the rest.
Consolidation with ultra-dense deployment
Let's look at an example situation where an ultra-dense deployment might make sense. Many companies have a development and test environment full of individual single-purpose servers. While these servers might be high-powered so that they perform well when compilation or tests are done, the rest of the time the servers sit idle. These idle servers not only waste power, they waste administrator resources whenever hardware fails. In this example, let's assume a development and test environment of forty servers on hardware a year or so from the end of its life cycle.
More on data center capacity planning
While this development environment might have been the latest and greatest back when it was purchased, it's likely that the specs are rather humble both in terms of CPU and RAM compared to today's top of the line. While specific numbers will vary, I've seen similar environments where up to ten of these mostly-idle machines could run on a dual-core, dual-processor blade if it had plenty of RAM, and even then it had some capacity left for failover.
In our example you could replace all forty of these development machines in just four blades. If you scale this idea to a fully-populated 10U blade chassis with 16 blades, you could potentially replace up to 160 1U servers.
Ultra-dense deployment to reduce power consumption
There are plenty of management benefits with high-density deployments, yet one of the first benefits is power usage. Many blade servers are designed to consolidate power consumption and even take extra steps to reduce the overall power draw, particularly when the system isn't being fully used. In our example, we have replaced 40 servers with 4. Even if the four servers use slightly more power than the older 1U machines, there is still a very significant power savings. Over time you could potentially pay for the hardware upgrade in power savings alone.
If you've been tempted by the growing push towards server consolidation, or if you have been bitten by the virtualization bug, now might be the time to consider both. Consolidation and virtualization are not just popular trends; when you combine them together you get a solution that saves management time, lowers upgrade costs, adds extra scalability to your environment, saves power, and might even pay for itself.
About the author:
Kyle Rankin is a systems administrator in the San Francisco Bay Area and the author of a number of books including Knoppix Hacks and Ubuntu Hacks for O'Reilly Media.
This was first published in April 2007