Evaluating adaptive routing in InfiniBand

InfiniBand is a state-of-the-art high performance networking technology used in many of the worlds largest supercomputers, clouds and datacenters where high bandwidth and low latency is required. The InfiniBand specification only supports static routing, but there are proprietary extensions for adaptive routing that has the potential to improve the performance and reliability of InfiniBand networks.

Adaptive routing is a feature that makes it possible to change the path that a network packet follows through the network. The actual path is selected based on the current network conditions. This can be used to improve network performance by distributing traffic across several links and always selecting the path with the lowest load. Adaptive routing is also beneficial for fault-tolerance where an alternative path can be used for communication when a switch or link fails. The challenges with adaptive routing are to understand when it is beneficial and how to make the right choice when you have several alternatives to choose from.

 

What you will do:

  • You will analyse and evaluate the adaptive routing features in InfiniBand.
  • You will make performance measurements on InfiniBand hardware.
  • You will calibrate our simulation model to InfiniBand hardware.
  • Potentially publish your results at a scientific venue!

What you will learn:

  • You will get to know the InfiniBand architecture, a state of the art high performance networking technology.
  • You will obtain a thorough understanding of adaptive routing, its benefits and drawbacks.
  • You will gain experience with the OpenSM software and the OFED stack.
  • You will learn to analyse network performance.

Qualifications:

  • Network experience
  • Programming experience

For more information, please contact Sven-Arne Reinemo (svenar@simula.no).

Emneord: Adaptive routing, interconnection networks, super computers, performance measurements, InfiniBand
Publisert 9. sep. 2016 11:44 - Sist endret 9. sep. 2016 11:44