Network Latencies

Description:

What are the driving forces behind network latency? What is latency?

Latency is the amount of time needed for a packet to go from place A to B, and to receive a response packet from B back to A.

[t=0, place=A] send packet to B ==> [t=25, place=B] Sent response to A ===> [t=50, place=A]

Network latency together with bandwidth are the two major harware inhibitors for parallel computing. This holds for clusters and grids, but also for system-on-a-chip solutions like the Cell architecture, which in fact is a cluster/grid on a chip.

Forces Stimulating the Development of Low Latency Infrastructure:

• Demand for latency sensitive applications:
  • Online computer games
  • high performance computing (Universities, Pharmaceutical, other R&D departments)
  • video conferencing
  • voice over IP
  • Lots of new market opportunities (High Performance Computing for everybody):
    • Complex Stock Market Analysis on Demand
    • Remote control of equipement X
    • Complex whatever, for consumers etc.

• Optical Fibres
• Increasingly faster integrated circuits (for network equipement)

Forces Inhibiting Low Latency Infrastructure Development:

• Speed of light
• Network Switching Techology
• Cost of (Re)placing new Switches and Cables
• Lack of business/consumer applications: Hard to exploit parallelism

Paradigms:

Speed of light ultimately limits network latency. To reach the other side of the world and back again (round-trip of 40.000 kilometres) with the speed of light, it would take us: 40.000/300.000 ~= 0.1333 seconds ~= 133 ms. Due to switching and other hardware extra delay is added and thus takes a little longer (280ms-350ms).

Timing:

The creation of the internet, and consumer/business interest into it enables faster development of high bandwidth and low latency networks.

Web Resources:

[1] Air Distances