Grids must be built "on top of" underlying hardware. This physical infrastructure of a grid -- things like computers and networks -- is often called the grid "fabric".

Networks are an essential piece of this underlying hardware. Networks link the different computers that form part of a grid, allowing them to be handled as one huge computer.

Networks are characterized by their size (local, national and international) and throughput(the amount of data transferred in a specific time). Throughput is measured in kbps (kilo bits per second), Mbps (M for mega; a million) or Gbps (G for giga; a billion).

One of the big ideas of grid computing is to take advantage of ultra-fast networks, whch allow the use of globally distributed resources in an integrated and data-intensive way. Ultra-fast networks also help to minimize latency: the delays that build up as data are transmitted over the Internet.

Grids testbeds are built on high-performance networks, such as the intra-European GEANT network or the UK SuperJanet network, which have 10Gbps performance on the network "backbone", which links the major "nodes" on the grid (like national computing centres).

One level down from the "backbone" are the network links, which join individual institutions to nodes on the backbone. Performance of these is typically 1Gbps. A further level down are the 10 to 100Mbps desktop-to-institution network links.

But what does "high-performance" mean?

Performance is measured in flops. A flop is a basic computational operation - like adding two numbers together. A Gigaflop is a billion flops, or a billion operations.

What more? Why not take a look at Internet Backbone Networks in the Atlas of cyberspaces. Or check out the Top500 high-performance networks and grid nodes.