Andrew Hillier, CTO and co-founder of CiRBA, spent some time discussing how a different view about datacenter usage would allow companies to reduce their costs, improve overall performance, and make it possible to better manage their datacenters. It is a pretty simple concept, but has some profound implications. Hillier suggested that datacenters be viewed like hotels, rather than like apartments.
Hotels versus apartments
Many companies view their datacenters like apartments. Systems were acquired to house a single set of workloads for a long period of time. Hillier asserted that today's modern, virtualized datacenters work differently. This, he pointed out, is the reason why performance and configuration management tools don't make the best use of available resources, and costs are higher than really needed.
Hillier suggested that hotels are a better model for how today's datacenters are being used. Virtual workloads come in, stay a while, and then leave. If resources aren't reclaimed and used to support a different virtual workload, datacenter efficiency and overall datacenter performance suffer. Costs for systems, storage, and software would all be higher than really necessary.
Making configuration management into a game like Tetris
IT administration would be well advised to use tools that make it possible to learn how these workloads run, what resources they use, and how they interact with one another, and then place them on systems. The process, he pointed out, is a bit like playing a game of Tetris with virtual workloads.
Deep analytics show where things should go
CiRBA uses what Hillier called "deep analytics" to learn how workloads work, what resources they use, and even the impact of plans for future expansion to transform how workloads are placed. CiRBA's technology can then automate the process of instructing systems, virtualization, and other management software to move workloads into an optimal configuration.
CiRBA can currently analyze workloads executing on Mainframes, midrange systems, and X86-based industry standard systems. It can then optimize workloads running on X86 and Power-based systems.
Viewing systems as if they were a hotel chain is a very simple model, offering some profound insights. Rather than thinking about workloads being static, moving in, and then staying in an apartment, it is rather helpful to think about virtualized environments being like a hotel. Travelers come in, stay a while, and then move on. The goal of hotel management is hosting the largest number of travelers in the most efficient manner.
If I consider today's evolving software licensing policies that allow multiple virtual insistences of a given product to execute on a single physical system using one software license, making as many virtual workloads as possible share one physical system could certainly reduce software licensing costs.
Since more "eggs would be in a single basket", it is very important that performance, resource utilization, and other operational characteristics be monitored very carefully. If performance anomalies start to appear, it would be wise to quickly find out what is happening, and move workloads to other "hotel rooms" quickly to prevent outages.
CiRBA's ideas are interesting and could be very valuable to companies having highly dynamic datacenters or operators of multi-tenant datacenters. If your company has this type of operation, learning more about CiRBA's technology could be very helpful.