Content area
Full Text
The Infiband architecture represents a giant step in the evolution of highperformance, switched-fabric interconnect systems. Its goal is to provide a highperformance, reliable, scalable way to connect high-end servers to other servers, I/O subsystems and the routers and switches that connect to the world outside the data center.
It does this by making possible the "dis-integration" of the traditional server. Today, each server is a complete system with CPU, memory, disk and I/O. Server farms are created by connecting servers to each other via Ethernet and to shared disk arrays via Fibre Channel.
Infiniband encourages the regrouping of those components so that a stack of thin CPU-only servers might occupy one rack, disks another and I/O backplanes still another-all connected to the Infiniband fabric. Instead of each server connecting to the Internet or Fibre Channel storage-area network via a dedicated PCI card, the Infiniband fabric would connect to a terabit router or Infiband to Fibre Channel bridge.
PCI devices would be located in a PCI-only backplane that also connects to the fabric. All resources on the fabric would be available to all components. This helps eliminate the dreaded "forklift upgrade" when new technology comes out. Instead of replacing an entire server (CPU, disk, I/O), just the needed components can be upgraded, and the entire fabric can take advantage of the increased performance. The server farm is now the fabric, and the fabric is the system.
By disintegrating the traditional server system, Infiniband changes the way servers must be validated. Instead of focusing on just one box and all the buses inside it, validation teams may need to simultaneously look inside several boxes as well as the fabric that connects them.
Dive mix of needs
Infiniband validation spans a huge range of technologies, from Microwave signaling to high-speed digital buses to protocol analysis and traffic generation across many layers of the Open Systems Interconnection model. Depending on the type of problems you find, you may need anything from 20-GHz timed domain reflectometer (TDR) scopes and 2.5GHz parallel bit-error-rate testers to logic analyzers, protocol analyzers and programmable traffic generators. A structured approach and a solid understanding...