Structured Data and Machine Learning for Enhanced Data Center Management

What is Structured Data?

Data centers produce a wealth of detailed information in their day-to-day operations. Making the best use of this data will help operators run a tight ship. One of the ways data centers can make the best use of this data is by having it in a structured data format. Structured data is a data set that is stored in a manner that allows you to group, filter and correlate that data in different ways. An important aspect of structured data that maximizes its value is normalization of values and format. For time-series data − such as real-time collected values − the time markers should also be normalized (for example, GMT), not localized, so the time stamps are uniform. Data accuracy here is critical; obviously, well-structured but inaccurate data is not valuable, as it translates to inaccurate analysis. This large amount of data includes details from top level power distribution gear like utility transformers and MVS (Medium Voltage Switchgear), power distribution like PDUs, RPPs, UPSs, as well as resource utilization, and environmental factors such as temperatures of rooms and racks. For some deployment this data even includes servers or CPUs.

The Value of Structured Data

Structured data is highly useful for several reasons, and its advantages include:

• Interoperability and ease of analysis
• Ease of storage and retrieval
• Data consistency
• Data validation and integrity
• Reporting and visualization

Let’s focus on interoperability and ease of analysis − well-structured, normalized data enables easier and more powerful processing and analysis of that data to identify patterns and spot trends.

DCIM Should Provide a Structured Data Set for Your Infrastructure

DCIM collects a significant amount of data, which by its nature should be stored in a well-structured format. This data is often used to analyze the infrastructure, historical events, and to drive real-time alarm notifications.  These requirements lead to a need for high quality of data, normalization, as well as accurate time stamping of the value.

DCIM Data is Ready-Made for Machine Learning

With a large data set, high accuracy, and time stamp tracking, the historical data in a DCIM is very well suited for a “tidy” data set essential for machine learning.  The historical data set can be used for model building and training − the basis of machine learning.  The larger the data set, the more accurate the models and results, thus the very large data set of a DCIM solution can produce high quality machine learning output.

Modius® OpenData® AI for anomaly detection

Modius has released its first phase of machine learning for OpenData AI– anomaly detection.  This module can create user defined models, with user selected devices, points, and time periods. OpenData AI can “self-train” from the historical data captured in OpenData. This allows the machine learning module to begin analyzing real-time almost immediately upon deployment. Anomaly detection can determine when real-time data does not align with the thousand or tens of thousands of past scenarios and determine when new conditions indicate a potential problem − well before a human could detect the same issue − and even before point values trigger manually defined alarms. In high uptime data center operations, machine learning brings awareness of potential issues before they become problematic, giving you an edge, and helping ensure optimal performance and protecting critical SLA agreements. OpenData AI also supports predictive and condition-based maintenance programs.