InfiniBand Network Analysis and Monitoring using OpenSM
Nishanth Dandapanthula · OhioLink ETD Center (Ohio Library and Information Network) · 2011
As InfiniBand (IB) clusters grow in size and scale, predicting the behavior of the IB network in terms of link usage and performance becomes an increasingly challenging task.The IB specification provides a detailed subnet management infrastructure to handle various aspects of the network.Open Subnet Manager (OpenSM) is an implementation of this specification and is available as a part of the Open Fabrics software package.Although many studies have identified the various subnet management phases, there exist no standard benchmarks that quantitatively identify the time spent in these phases and analyze the functional intricacies of OpenSM.There currently exists no open source tool that allows users to dynamically analyze and visualize the communication pattern and link usage on the IB network.In this thesis , we design a set of micro-benchmarks using OpenSM to study the various phases of subnet management at a finer granularity.We design and develop a scalable InfiniBand Network Analysis and Monitoring tool -INAM.INAM monitors IB clusters in real time by querying various subnet management entities in the network.It is also capable of capturing the pattern of traffic on a subset of links in the network.This allows users to visualize and analyze the network communication characteristics of a job in a high performance computing environment.We demonstrate the effectiveness of INAM using a wide set of micro benchmarks and applications.