A place for the occult in the data centre?
Fri 22 Jul 2016
I had not realised quite how parochial component manufacturers can be until I read recently of the work that researchers from Yahoo have been doing, in concert with Carolina’s Duke University, to create a framework for unlocking useful data from the syslog outputs of network devices, in order to determine operational issues hidden in the millions of opaque messages they output every day in even a medium-size DC operation.
The devices in question gabble on day-long about themselves, and their entire output is only useful for analysing problems that they are having…with themselves. As the researchers note, turning this verbose stream of inward-looking information into an open resource would, in many cases, involve reverse-engineering proprietary systems, in order to structure that information into a useful context.
Instead the researchers have employed a Quasi-Experimental Design (QED) system to infer causal relationships between what the syslogs say and what other devices report. Apparently the system, entitled Log-prophet, works well, distilling millions of lines of un-timestamped and unstructured text into a small and reliable number of distinct DC events.
On the one hand Log-prophet is an inventive repurposing of information – but on the other hand, it’s a convincing argument for open-source hardware frameworks with configurable and flexible levels of feedback, built with the understanding that no hardware is ever going to operate in a vacuum.
It is a sorry state of affairs when researchers have to experiment with social science methodologies – more normally used to determine trends in Twitterstorms – in order to make data centre equipment susceptible to general monitoring schemes. In fact, it’s practically in the realms of the occult.