A few weeks ago I had the pleasure of hosting the machine data track of talks at Strata Santa Clara. Like “big data”, the phrase “machine data” is associated with multiple (sometimes conflicting) definitions, two prominent ones come from Curt Monash and Daniel Abadi. The focus of the machine data track is on data which is generated and/or collected automatically by machines. This includes software logs and sensor measurements from systems as varied as mobile phones, airplane engines, and data centers. The concept is closely related to the “internet of things”, which refers to the trend of increasing connectivity and instrumentation in existing devices, like home thermostats.
More data, more problems
This data can be useful for the early detection of operational problems or the discovery of opportunities for improved efficiency. However, the decoupling of data generation and collection from human action means that the volume of machine data can grow at machine scales (i.e., Moore’s Law), an issue raised by both Monash and Abadi. This explosive growth rate amplifies existing challenges associated with “big data.” In particular, two common motifs among the talks at Strata were the difficulties around:
- mechanics: the technical details of data collection, storage, and analysis
- semantics: extracting understandable and actionable information from the data deluge
The talks covered applications involving machine data from both physical systems (e.g., cars) and computer systems, and highlighted the growing fuzziness of the distinction between the two categories.
Steven Gustafson and Parag Goradia of GE discussed the “industrial internet” of sensors monitoring heavy equipment such as airplane engines or manufacturing machinery. One anecdotal data point was that a single gas turbine sensor can generate 500 GB of data per day. Because of the physical scale of these applications, using data to drive even small efficiency improvements can have enormous impacts (e.g., in amounts of jet fuel saved).
Moving from energy generation to distribution, Brett Sargent of LumaSense Technologies presented a startling perspective on the state of the power grid in the United States, stating that the average age of an electrical distribution substation in the United States is over 50 years, while its intended lifetime was only 40 years. His talk discussed remote sensing and data analysis for monitoring and troubleshooting this critical infrastructure.
Ian Huston, Alexander Kagoshima, and Noelle Sio from Pivotal presented analyses of traffic data. The talk revealed both common-sense (traffic moves more slowly during rush hour) and ￼￼￼￼￼￼￼￼￼counterintuitive (disruptions in London tended to resolve more quickly when it was raining) findings.
My presentation showed how we apply machine learning at Sumo Logic to help users navigate machine log data (e.g., software logs). The talk emphasized the effectiveness of combining human guidance with machine learning algorithms.
Krishna Raj Raja and Balaji Parimi of Cloudphysics discussed how machine data can be applied to problems in data center management. One very interesting idea was to use data and modeling to predict how different configuration changes would affect data center performance.
The amount of data available for analysis is exploding, and we are still in the very early days of discovering how to best make use of it. It was great to hear about different application domains and novel techniques, and to discuss strategies and design patterns for getting the most out of data.