Unsupervised fault detection for Cloud Platforms

5G (ENCQOR) Technology Development Challenge

Unsupervised fault detection for Cloud Platforms

Challenge Launch Date

May 24, 2019

Challenge Deadline 

September 6, 2019 (This call for projects has expired. Notices of interest are no longer accepted.)

Challenge Statement 

Cloud platforms make up the backbone of 5G: they host most software functions of 5G. These platforms, including the (5G) software that runs on them, make a very complex system that that can fail in several different ways. The objective of this project is to explore the use unsupervised learning techniques for fault and anomaly detection in complex software systems, using the cloud platform itself as a use case.

Project Partner

Ericsson Canada Inc.

Timeline

2 years

Available funding

$60,000 ($30,000/year for the two Ph.D.)

Applicant Type

Quebec/Ontario based University (possibly selected through an RFP)

Location

In University’s lab and/or Ericsson Montreal site

Project Details

The worst way to detect faults is when service interruption occurs and end-users complain about it. To detect faults and anomalies in a timely manner, a common technique is to monitor the system for known symptoms faults. This however requires that a person with a deep knowledge of the system identify how to monitor for these symptoms. This approach, however, is not efficient since it depends on the knowledge of human expert to identify the faults.

The goal of this project is to develop a technique for unsupervised (or semi-supervised) fault and anomaly detection in a cloud platform.

• Use different sources of data, including performance monitoring and logs
• Develop techniques for unsupervised anomaly detection, and extend that to fault detection using semi-supervised approaches
• Identify methods for generalizing models learned at one site (e.g., a test deployment or lab environment) to another (e.g., live production system)

Project Goals/Outcomes

The deliverables for this project include:

• A fault and anomaly detection system that fulfils the above requirements
• Papers and patents documenting the results
• PoC implementation for a few cloud platforms (e.g., OpenStack, K8S, etc.)

Applicant Capabilities

2 researchers Ph.D. level with:

• Demonstrable expertise in machine learning and cloud platforms
• Good track record of publishing papers and writing patents
• Experience with unsupervised/semi-supervised learning, time series analysis, log analysis, system monitoring is a plus

Additional Information

• N/A