Prof. Kemme and PhD student Mona ElSaadawy are looking for a student interested in a NSERC/SURA undergraduate research internship. Interested students should have ideally taken already COMP 310.

Topic: Application Monitoring as a Network Service

The recent rise of cloud applications, representing large complex modern distributed services, has made performance monitoring a major issue and a critical process for both cloud providers and cloud customers. Many different monitoring techniques are used for such applications, from simply measuring resource consumption, to application-specific measures which instrument the application or platform, to analyzing message exchanges. In all these cases, the information to be analyzed is logged at the host nodes on which the application is deployed, and from there either analyzed on site or forwarded to analyzing tools. Using new trends such as Software Defined Networking (SDN) shows promise to move some of the logging functionality into the network as a lot of information can be extracted from messages exchanged between application components. This raises the potential for the cloud infrastructure to provide a Monitoring as a Service to cloud applications in a transparent manner avoiding software instrumentation and allowing for a more flexible placement of logging functionality. This SDN based approach uses techniques such as port mirroring and/or tunneling to enable message filtering and reformatting, and uses a customized port sniffer for software switches (such as OVS). We are currently designing the architecture of the overall Monitoring Service. As first prototype, the system should allow users to choose the entities that should be monitored in real-time (by entering IPs and/or port numbers), and select the monitoring matrices of interest, and the monitoring interval and duration. Our first prototype will target request/reply type protocols (HTTP, database, memcache, etc.) and should provide multiple monitoring matrices such as:

Request service time (minimum/maximum/average).
Throughput in terms of sent bytes and/or packets (minimum/maximum/average)
Error rate (the number of the failed requests out of all sent requests classified by the request type).
Host load (number of handled requests/all received requests).
Object popularity (the number of requests for a certain object(file, video, gif,...etc) / all received requests).
Possibly deep inspect (e.g., to detect certain words or sentences in the message).

The user interface should be a dashboard where the user can configure what and where to monitor, and monitor results should be visualized to the user in real time. In addition, our service should raise alarm in case of outliers in the monitored matrices. For example, if a certain application request service time, error rate, or host load exceeds a certain threshold (which may be an input from the service user or learned from the monitored application through machine learning techniques), the user should be notified. During the undergraduate research internship, the student is expected to actively participate in the design and the implementation of some of the component of this prototype.

If you are interested, please email to kemme@cs.mcgill.ca and mona.elsaadawy@mail.mcgill.ca