Version 1 (modified by, 13 years ago) (diff)


Internet Scale Overlay Hosting

The experimental system prototype for implementing the Internet Scale Overlay Hosting services is named Supercharged PlanetLab Platform(SPP). Information reviewed for SPP and its tutorial is available at:

SPP can be accessed at the Washington University's SPP PlanetLab Site.

Internet Scale Overlay Hosting Findings

SPP is a high performance substitute for a typical PlanetLab node, which normally is a PC running a customized Linux running multiple virtual machines, using a Linux vServer mechanism. The SPP design modifies this definition to achieve these goals:

  • Increase performance
  • Scale by incorporating multiple servers
  • Use Network Processor blades
  • Improve control of computing and networking resources

Both hardware and software components have been used to implement SPP, a high level outline is given below:

=> SPP Hardware Components

The system consists of a number of processing components that are connected by an Ethernet switching layer. The most significant parts of the system are the 2 types of Processing Engines:

  • General Purpose Processing Engine (GPE) - A dual processor blade (Radisys ATCA 4310) with 2 network GbE interfaces.
  • Network Processing Engine (NPE) - A Radisys ATCA 7010 blade with 2 Intel IXP 2850 Network Processors subsystems. NPE supports fast path processing and provides up to 10 Gb/s of IO bandwidth.
  • Line Card (LC) is used to handle all input and output.
  • Switching Substrate - A chassis switch (Radisys ATCA 2210) with 2 switches (1 fabric switch w/10 GbE ports, 1 base switch w/1 GbE ports)
  • ATCA Chassis - A Zephyr Shroff 5U six slot ATCA chassis to force component reboot.
  • Control Processor - a Dell PowerEdge 860 that allows maintenance access to base/fabric switches, serial interfaces, chassis switch blade, and the GPEs.
  • NetFPGA - a Xilinx Virtex 2 Pro 50 FPGA PCI card with 4 network connections available for forwarding.

=> Network Processor Datapath Software

Various software components use the Line Card to process incoming and outgoing traffic. One Line Card Network Processor is used to process incoming traffic and one s used to process outgoing traffic. The packet processing is structured as a pipeline where packets flow across stages which use one or more Micro Engine (ME). A detailed description is available at SPP Data Path Software.

=> Control Software

The central component of the system is the System Resource Manager(SRM). The SRM retrieves slice descriptions from PlanetLab Central (PLC), creates the slices, and allocates slice resources. A detailed description of the control software is available at SPP Control Software.

Note: Milestone SPP.S2.a geniwrapper, due 03/31/10 is late. Not sure if requesting access to the web site listed makes sense before geniwrapper is delivered.