Data analytics are becoming more sophisticated and more meaningful every quarter, which means that they are becoming more valuable for data center operations. As data centers evolve and embrace new connections and new connection strategies such as virtualization, it becomes more difficult to determine what impact these changes have on network performance. Using analytics, IT managers can identify and even predict potential choke points and errors before they create problems.
Just as companies are harnessing big data in order to address business problems, organizations are using big data analytics to look inward at the data center, uncovering unseen correlations between platforms, the impact of new workloads and the effect of user resources.
Applying Big Data to Data Center Operations
Big data systems typically run on virtual data centers, which means the data center itself is using various stacks of components. That means statistical and performance data is available from multiple network components that can be used for analytics.
Every network component—switches, bridges, storage systems, etc.—is logging data; however, the data format and criteria were developed by each component vendor, which makes it more difficult to correlate network data. What makes virtual data center monitoring and management more complex is that most of these network components are generating data deemed relevant before virtualization existed, so the metrics they provide aren’t as valuable for measuring virtual infrastructure performance. This makes it harder to distinguish cause from effect in a virtual infrastructure.
The data provided by virtualization itself can be invaluable in analyzing virtual data centers. For example, the hypervisor has a lot of information because it’s designed to use context-rich data to allocate virtual resources. Extracting hypervisor data using analytics allows you to map workload patterns and starts to provide a framework for identifying underlying system correlations.
Using data from the hypervisor, you can optimize workloads, identify new systems that can host additional workloads, and track changes over time. By creating a baseline for performance using the hypervisor data, you can also monitor cluster performance, rather than having to manage each host or virtual machine separately.
So just as virtualization gives you holistic control over network resources, it also gives you centralized analytics to help you fine-tune network performance. Using big data, you can identify weak points in the system and determine what changes might improve performance. For example, what are the performance differences when running the same virtual machine on different host systems from different vendors (e.g., Cisco, Dell or Hewlett Packard Enterprise)? The analytics show the impact on scaling the system, which is what you need to know to manage the data center effectively.
Next Step: Predictive Analytics
Once you have the means to gather and analyze data center performance metrics, you have the tools for predictive analytics. Using the end-to-end view of the enterprise infrastructure, performance data from the hypervisors, historical data and other metrics, you can model points of failure and predict where problems are likely to occur before they actually happen.
IBM and Juniper, for example, have already announced a partnership to develop real-time network analytics to support mobile applications and accommodate new demands, such as streaming Internet of Things applications. The objective is to assimilate enough data to allow IT managers to optimize performance on a predictive basis, rather than reacting to performance problems.
Analytics Inform SDN
One of the promises of big data is providing the intelligence to power the self-healing network. Real-time network analytics combined with predictive analytics makes it possible for data centers to manage themselves, isolating faults, rerouting data traffic and performing proactive management tasks without human intervention.
Using analytics to gain a holistic view of the network also lends itself to centralized network control using software-defined networking (SDN). The analytics inform SDN actions so the system can diagnose and repair network problems in real time.
Ultimately, next-generation administration platforms will be able to track real-time changes to the network on a per-application or per-user basis. That real-time data will power predictive models that anticipate problems likely to affect performance or use experience. IT managers will be able to automate remediation and get ahead of network performance problems by using analytics to automate SDN architectures.
Just as analytics reveal business trends, big data analytics can reveal a variety of performance and security anomalies in the data center itself. It also provides the means to automate remediation, using big data analytics to anticipate potential problems and instruct the system to take action in advance. These benefits give IT managers more incentive than ever before to invest in big data, which means they’ll soon be looking for more enterprise and cloud resources to deliver analytics.