交换机和路由器的特征
Characteristics of Switches and Routers
Overview
The purpose of this document is to explain the differences between Switches and Routers. A
key assumption made in this paper is that Switches are Ethernet oriented devices, rather than
ATM, Frame Relay or voice, and that they run TCP/IP based protocols rather than legacy
enterprise class protocols such as IPX. This document also provides guidance about where to
deploy each category of device, based on real operational experiences.
Executive Summary
In order to understand the differences between a Switch and a Router, we must first
understand the environment in which they operate. A Switch was designed to operate in a
Local Area Network, while a Router was designed to operate in a Wide Area Network. As
such, the platforms’ fundamental design parameters address different sets of requirements in
terms of software capabilities, function optimization, dependability, performance, scale and
density. Traffic patterns and demands are very different in each environment, and an
approximate rule can be applied: LAN’s have 80% local traffic and 20% external traffic,
compared to WANs which reverse this trend. In addition, LAN traffic patterns can be flow
oriented and show a degree of predictability. For example, FTP servers, Intranet servers and
printer resources. This is in contrast to the largest WAN, the public Internet, where networks
can be reachable via several different routes, traffic can be asymmetric in nature by returning
down a different path, and traffic patterns and quantities constantly change. Another way to
view this is that LAN traffic flows tend to be n to 1, and WAN traffic flows tend to be n to n.
The 80/20 rule should be regarded as an approximate guide because new disruptive
technologies, applications and organizations are constantly appearing. For example the peer-to-peer traffic driven by applications such as Skype and e-Donkey. These applications echo
the origins of the ARPAnet in the pre-Internet era, which carried peer-to-peer ambitions.
Layer 2 Switches are in effect a set of learning or filtering bridges internally interconnected,
which are capable of traffic reductions and loop prevention in an LAN. A Layer 2 Switch has
virtually no use in a WAN because of limitations with respect to scalability, flexibility,
efficiency and security. Superficially, a Layer 3 Switch looks similar to a Router. However, the
Switch was conceived and designed (and architecturally optimized for) for LAN
environments, which leads to shortcomings in WAN environments. The key areas of shortfall
are: Lack of Delay-bandwidth-buffering, limited congestion buffering and traffic control,
inferior routing algorithm robustness, limited IP service creation capability and a smaller
range of interfaces.
In this paper, we examine Switch and Router capabilities, and the needs of LAN and WAN
environments. By understanding these requirements we can conclude which devices are most
suitable in each environment. The lower cost of a Layer 3 Switch can then be compared to the
features and functions of a Router, so that a properly informed deployment decision can be
made. Introduction
An industry-wide debate has been in progress for a considerable length of time focusing on
the ability of Switches to perform the functions normally associated with Routers in a much
cheaper and simpler form. This is potentially true for the Internet and WANs, and also other
topological areas such as Metro Area Networks (MANs), and “Last Mile” or access nodes. The
wide distribution of Ethernet technology has added fuel to the debate, and in fact many
service providers are now choosing Metro Ethernet Switches in their networks.
It is a commonly held belief that by deploying Ethernet, the cost of service provider operating
expenses will be driven downwards and will be beneficial to both providers and their
customers. This assumption is reasonable, given that history shows Ethernet port costs have
consistently been driven down in Enterprise and Local Area Networks (LAN). However,
applying LAN technology into a WAN environment is not a trivial matter, especially the issue
of scale found in a WAN the size of the Internet. Ethernet has physical limitations such as a
maximum of 1024 nodes in a single collision domain, and although an Ethernet network can
be expanded far beyond this by using repeaters and Switches, it is impractical on a very large
scale. Hierarchical network design is an essential scalability requirement, which is why IP is a
suitable technology for global communications. Despite the issue of scale, point-to-point (full
duplex) Ethernet access pipes are proving to be an effective means of delivery layer 2 services.
Before Layer 3 routing dominated data networking, Layer 2 bridged WANs were common
place. Many Layer 2 technologies became available, such as Token Ring, FDDI and LANE, but
over time Ethernet in its various forms emerged as the dominant technology of choice,
primarily in LANs. Ethernet has evolved from 10Mbps to the current 10Gbps developments,
and is even taking on SONET/SDH like attributes, as demonstrated by the IEEE’s 802.17
Resilient Packet Ring (RPR) technology. Meanwhile, the Internet, which is based upon routing IPv4 protocols, has comprehensively out-scaled every Layer 2 based infrastructure. However
Layer 2 is still very important to service provider networks, and is trusted to provide multi-
service connectivity, for example Voice over Frame Relay and ATM. In the future, Layer 2
based MPLS VPNs will emerge as a key revenue generating technology, extending the
benefits of more traditional methods over a converged IP/MPLS infrastructure. These services
are also useful when transporting legacy customer’s protocols, rather than IP such as SNA,
IPX, DECNet and Appletalk, which are still used in specialist applications.
Many Ethernet Switches now have Layer 3 capabilities, which feature IP as the dominant
protocol. More than ever before, service providers are now under pressure to reduce
operating expenses, while offering new IP based services. As Layer 3 Switches are generally
cheaper than Routers, many service providers are also asking: Why shouldn’t Layer 3
Switches be used in the place of Routers? This issue is addressed in the following sections on
Layer 3 platforms, but the key areas of difference between Switches and Routers can be summarized in five general areas: Scalability, Reliability, Features, Management and
provisioning issues, Cost.
Layer 2 Switches
At a point in the 1990’s, the term “Layer 2 Switch” began to appear in the data networking
industry. It was a term used to describe a LAN platform, which was capable of extremely
high performance frame forwarding based on MAC layer addresses. If this sounds familiar, it
is because this is the basic operation of a Bridge, but with higher performance.
When most people refer to a Bridge, they really mean a “Filtering Bridge” or “Learning
Bridge”. For the purposes of this document, the term Bridge (or Layer 2 Switch) is referred to
in this context. A Bridge relays every frame received on any port to every other connected
LAN. If a frame is received on the same LAN as its destination, then it is not forwarded to
other LANs. As they only forward broadcast and non-local traffic, traffic reductions through
the LAN topology can be achieved, which also allows reasonable performance over slow
WAN links. These devices are designed to examine Layer 2 information such as the Media
Access Control (MAC) source and destination address and act upon it. Layer 3 Switches and
Routers examine more information further up the ISO 7 layer model, including IP source and
destination address, and act upon it.
2001-2007 Comsenz Inc.
