The GEC4 Demo

From ARL Wiki
Revision as of 20:44, 13 May 2009 by WikiSysop (talk | contribs)
Jump to navigationJump to search

The GEC 4 Demo <ref name="The fourth GENI Engineering Conference Demonstration"> given in early 2009 focused on the fastpath features of the SPP. That is, an SPP node supported both multiple physical interfaces and multiple slices, and an SPP slice has direct control of resource allocation. In particular, a slice can:

  • Reserve a share of each interface's maximum bandwidth
  • Bind and dynamically configure (size and bandwidth share) multiple packet scheduling queues to each interface
  • Install filters that forward packets to specific packet queues and output interfaces

This page describes how to do the Demo. This description will highlight the features listed above as well as provide an example of some details that will be used in any use of the SPP.

Before describing the Demo, we briefly review the main SPP concepts as an aid to understanding the GEC 4 Demo. The SPP is a PlanetLab node that combines the high-performance of Network Processors (NPs) with the programmability of general-purpose processors. This means that most data packets should be handled by the fastpath provided by NPs while control and exception packets are handled by the slowpath. In order to provide this service dichotomy for an application, an SPP user does the following for each of its slices:

  • Allocate (and configure) a fastpath
  • Create, configure and bind packet queues to endpoints of one or more UDP tunnels
  • Install filters that indicate where packets should be forwarded (routed)

Overview of the GEC 4 Demo

Configuration figure The configuration contains three routers (R1, R2 and R3) and six hosts in which R1, R2 and the six hosts form a dumbbell configuration and R3 is a transit router that is attached to both R1 and R2. All links going to a host have a capacity of 150 Mb/s, and all interrouter links have a capacity of 300 Mb/s.

Part 1: Basic Demonstration

Part 2: Effect of Queue Parameter Changes

Part 3: Effect of Filter Changes

Conclusion

References