Computer Networks Basics

• The overall performance of a cluster system can be determined by
  1. the speed of its processors and
  2. the interconnection network.
• Many researchers argue that the interconnection network is the most important factor that affects cluster performance.
• Regardless of how fast the processors are, communication among processors, and hence scalability of applications, will always be bounded by the network bandwidth and latency.
  • Bandwidth is an indication of how fast a data transfer may occur from a sender to a receiver.
  • Latency is the time needed to send a minimal size message from a sender to a receiver.

 $$

Networks can be divided into the following four categories based on their sizes and the geographic distances they cover:

1

Wide area network (WAN); a WAN connects a large number of computers that are spread over large geographic distances. It can span sites in multiple cities, countries, and continents.

2

Metropolitan area network (MAN); the MAN is an intermediate level between the LAN and WAN and can perhaps span a single city.

3

Local area network (LAN); a LAN connects a small number of computers in a small area within a building or campus.

4

System or storage area network (SAN); a SAN connects computers or storage devices to make a single system.

• In the early days of clusters, Ethernet was the main interconnection network used to connect nodes.

Table 1: Data Rate, Switching Method, and Routing Scheme for Interconnection Networks.

Interconnection	Switching	Routing
Ethernet	Packet	Table-based
Fast Ethernet	Packet	Table-based
Gigabit Ethernet	Packet	Table-based
Myrinet	Wormhole	Source-path
Quadrics	Wormhole	Source-path
Infiniband	Packet	Source-path

• While Ethernet resides at the low end of the performance spectrum, it is considered a low-cost solution.
• Other solutions add communication processors on the network interface cards, which provide programmability and performance.
• Table 1 shows the relative performance and other features of different high-speed networks.

• The major factor that distinguishes WAN from other network types is the scalability factor.
• LAN technologies provide higher speed connections compared to WAN because they cover short distances and hence offer lower delay than WANs.
• Network routing schemes can be classified as
• connection-oriented; in connection-oriented, the entire message follows the same path from source to destination.
  • Only the first packet holds routing information such as the destination address.
• connectionless; in connectionless schemes, a message is divided into packets.
  • The packets of a given message may take different routes from source to destination.
  • Therefore, the header of every packet holds routing information.
  • Using a serial number, the message can be reassembled in the correct order at the destination as packets may arrive in a different order.