This lecture is divided into hyperlinked sections
LAN switch fundamentals
Advantages of
Switched LANs
Switched Ethernet
Types
of Ethernet Switches available
Gigabit Ethernet
Switches
Switches
bridges and routers compared
Layer 2 Switches
Layer 3 Switches
LAN switch architectures
Adaptive switches
Virtual LANs
Layer 2 Virtual LANs
Port Grouping
MAC address Grouping
Layer 3 Virtual LANs
Summary of VLAN types
Other Switched LANs
Migration
to Switched Token Ring
FDDI switching
Conclusion
What is the motivation for LAN switches? With Ethernet
LANs, 12 users sharing 10 Mbit/sec get share of 40% of the network’s bandwidth.
With an eight-port switch replacing the bus, 12 users will benefit from
the addition of the switch giving up to
40 Mbit/sec concurrently to the users.
The network usage may be calculated by
Number of switch ports
x data rate
2
Figure 6.1 Twelve Users on shared medium LAN
Figure 6.2 Twelve Users benefit from 8-port switch
In a non-switched LAN, stations have the use of shared bandwidth and each station gets all of the bandwidth some of the time. Stations contend for access time (CSMA/ CD, token, demand priority). As the user numbers on a given network grow, each user gets a smaller share of the available bandwidth and the increased traffic brings its own problems of access delay and collisions. The solution is to replace the shared-bandwidth media with a switch giving each switch port full bandwidth. The switch can service many ports simultaneously and as the number of ports is increased so the bandwidth grows with port delay remaining very low and constant. This is a very cost effective solution and uses existing cabling NICs and applications.
One advantage of a switched LAN is that it divides the network into small independent fragments and interconnects them as needed. The internal data connection rate is very high. It effectively multiplies the LAN bandwidth by half the number of ports with the ability to use wire-speed ports simultaneously. The 8 port Etherswitch in figure 6.2 offers users 40 Mbit/sec on demand. Switches are easy to deploy into an existing network in phases and are scalable upward in port numbers, but can make for a blocking network.
Switched Ethernet can remove bottlenecks from existing Ethernet segments and users and applications that demand more bandwidth can be satisfied. It retains investment by using existing hardware such as adapters, stations and servers and also makes use of existing software, NOS, drivers and applications. It is a safe cost effective reliable proven technology and can be phased in whenever it is needed. It also will co-exist with shared Ethernet
Figure 6.3 (a) Bus-type Ethernet
Figure 6.3 (b) Switched Ethernet LAN
Shared medium Ethernet has only 10 Mbit/ sec shared capacity with just one collision domain. With a large network there are maintenance and administrative problems. An 8-port switch can multiply existing bandwidth from 10 to 40 Mbit/ sec.
Full duplex Ethernet is now available, using different pairs of UTP for separate signal paths. This means that ports having one station or server attached can enjoy simultaneous transmission and reception. This would probably require an adapter and NOS upgrade. Additional benefits of switched Ethernet include:
· Co-location of switches and servers in a place
of safety
· Use of existing cabling
· It should be possible to use existing hubs and
bridges.
Types of Ethernet Switches available
Many vendors offer a wide range of switches in this area, offering cut-through, store-and-forward and adaptive. There are different sizes available too, workgroup, departmental and backbone.
· Fragment-free switches may be purchased and these
have the capability of removing left-overs from collisions, i.e. runt frames.
This is done by storing and examining the first 64 bytes of a frame.
· 10/ 100 switches can be either 10BASE-T with
one or two 100 Mbit/ sec ports or with 10/ 100 ports.
· Fast Ethernet switches often offer optional
FDDI or ATM backbone ports. Bridging modules are often built in too.
Gigabit Ethernet switching is available, the first products being targeted at the LAN backbone market. Many high-end Fast Ethernet switches offer Gigabit Ethernet uplinks.
When switched, Gigabit Ethernet can operate even more like other Ethernet:
· Full duplex can be used on ports with a single
MAC
· CSMA/ CD can be discarded
· Carrier extension is no longer required
· Frames can be identical to other Ethernets
Figure 6.4 Switched Gigabit Ethernet Implementation
Figure 6.5 Deployment of Ethernet Switches throughout an Organisation
Switches bridges and routers compared
A switch may be envisaged as a fast multiport bridge. Switches may be grouped by the protocol layer that they process. The example in figure 6.6 below shows an IP packet that has been encapsulated in an Ethernet frame.
Figure 6.6 IP packet Encapsulation
Layer 2 switches make routing decisions on the information provided by the Data Link layer. The forwarding decisions are based on MAC addresses contained within the Ethernet frame header. A layer 2 switch is essentially a very fast multi-port bridge.
The advantages of layer 2 switching include:
· Protocol independence allowing switching of IP,
IPX, Appletalk etc.
· Non-routable protocols e.g. NetBEUI may be switched
· Very low delay connecting end points
· Least expensive upgrade option
· Multivendor interoperability is possible allowing
easy integration of equipment
Layer 2 switches can be easily deployed where repeating hubs and bridges have previously been deployed by simply removing the old equipment and replacing with the new switches.
Layer 3 switches make routing decisions based on network layer information with forwarding decisions based on addresses in the packet header. These are very fast multiport routers. Advantages of layer 3 switches include:
· Utilisation of Network Layer subnets allowing
isolation of network traffic within the subnets thereby providing better
security
· Switches maintain topology information i.e.
routing tables
· Large networks may be built up easily i.e. good
scalability
Disadvantages of layer 3 switches include:
· They are generally very expensive
· Apart from a few exceptions, multivendor interoperability
is not yet possible, vendors currently use proprietary solutions for exchange
of routing information
Layer 3 switches are generally deployed where routers have been previously used. They avoid network hops by direct routing giving shortcuts through the networks to the destination. Router decision times are speeding up, the previous software based algorithms are being replaced by ASICs (Application-specific ICs) providing a hardware solution (faster than software).
The switches are able to perform store-and-forward bridge-like functions where:
· The entire input frame is stored in memory
· FCS checked, errored frames being discarded
· Unerrored frames are read out of memory to the
proper output port
Because the received frames are checked for errors, no errored frames will be forwarded, but the checking produces longer throughput-delay and also the switches need a large amount of memory to cope with network traffic fluctuations.
Figure 6.7 Operation of level 3 switching
Another type of switch available is the cut-through switch, characterised by lower latency than the store-and-forward type. Frame error control is not performed, as soon as the frame has arrived its destination address is processed and the frame will be forwarded towards the destination (provided the medium is free) else output buffering will be required. Advantages of cut through switching include very little buffering and little delay through the switch. Disadvantages are that errored frames will be forwarded to their destinations and frames may be dropped if there is insufficient frame buffering at the output and the destination medium is busy.
Fig 6.8 Operation of a Cut-through switch
Adaptive switches have the highest per-port cost but manage to combine the best features of store-and-forward and cut-through switches. During operation the switches default to cut-through switching but the FCS is checked as the frame leaves the output port. The number of errors are counted and error statistics are associated with the corresponding input port. If the errored statistics rise above some pre-determined level the affected input port switches are returned to store-and-forward operation. The errored-frame threshold may be set by the network manager or alternatively use may be made of the vendor default value.
Figure 6.9 Deployment of Switches throughout an Organisation
In figure 6.9 each switch must keep a record of the MAC addresses connected to each port. The backbone switch potentially has to keep a record of hundreds of MAC addresses per port.
The workgroup switches bring switching benefits to the desktops of the power users. Each switch will be a small stand-alone item with 8 to 16 ports. These switches have limited management features and may only be able to store one MAC address per port.
Departmental switches have a higher level of sophistication and can switch several LAN segments. They are often used in horizontal applications i.e. on one floor of a building. They are often stackable and so easy to upgrade with management features being available as an option.
Backbone switches often incorporating routing functions are a replacement for the LAN backbone. They connect departmental switches and have a chassis with modular plug-ins. They are fully manageable and often contain software for traffic monitoring (RMON). They can route between different VLANs.
Read more about switched LANs.
A virtual LAN (VLAN) is a logical grouping of stations defining arbitrary users and resources that may be some distance apart and otherwise would not normally be part of the same LAN.
Figure 6.10 Layout of a Virtual LAN
The port settings within the LAN switches define the extent of and stations within the VLAN. A station within VLAN 1 communicating with another station within VLAN 1 will route its message via the VLAN switch using ports A, B or C. The frames will not be repeated on ports D, E or F. Several stations may be connected to the same port and therefore must be part of the same VLAN.
In the past network administrators have physically segmented LANs using segments that have been hard-wired in place and these segments would typically be connected by bridges and routers. Any moves or additions would have mean physical changes in wiring or location of the station.
The VLAN offers logical segmentation with the network administrator defining arbitrary groups of users. All resources and traffic and broadcasts remain within these defined groups. Physical moves and cabling alterations are unnecessary, thus a user may be moved from VLAN 1 to VLAN 2 in software by changing port assignments.
VLANs promise:
· Easier moves, additions and changes to logical
structures
· Enhanced collaboration between stations within
the same VLAN
· Better security by isolation of traffic
Layer 2 switches group virtual LANs using information provided by Data Link layer information, the VLANs being interconnected using bridge-like functions within the switches. VLAN members may potentially be scattered world-wide. It is possible to administer the switches remotely using TELNET, each switch having its own IP address.
Advantages of layer 2 LANs include:
· Protocol independence e.g. IP, IPX, NetBIOS etc.
· Low latency by provision of cut-through switch
operation
Disadvantages include:
· Limited scalability, only good up to moderate
sized networks
· Proprietary switch-to-switch signalling, switches
connected using 802.Q. A recent innovation is 802.10 which hopes to cover
links between different vendors' switches.
There are two methods of defining layer 2 LANs, either switch port groupings or MAC address groupings.
This is the least sophisticated method of VLAN grouping and the assignments and any changes may be performed by the network administrator. It is the least expensive method of grouping.
Figure 6.11 VLAN formation by Port Grouping
MAC address grouping is more sophisticated. Each switch has its own routing table and if a station is physically moved, the switch will automatically learn the new location and update the port tables. Here several VLANs may be combined into the same port but there is a security disadvantage insofar as stations from one VLAN may see frames destined for a different VLAN that have been assigned the same port number.
Figure 6.12 VLAN formed by MAC address grouping
Layer 3 switches group VLANs using information from the network layer. VLANs are interconnected using router-like functions within the switches. VLANs are identified by different subnet numbers and the groups can be identified by use of the same protocol e.g. all IP, all IPX.
Advantages of layer 3 VLANs:
· There is more security with sophisticated filtering
and it is possible to create multiple firewalls.
· There are multiple subnets possible per port.
e.g. engineering and manufacturing subnets may overlap.
Disadvantages of layer 3 VLANs:
· Higher latency caused by examination of packet
to determine routing means store-and-forward type switching must be employed.
The time-to-live is decremented at each router.
· Sophisticated functions of conventional routers
are not provided e.g. protocol conversion.
Virtual networking is still in its early stages of development with many proprietary solution existing but true interoperability is still some way off.
Some issues worth considering when discussing VLANs with potential vendors are:
· Network cost
· Ease of configuration and reconsideration e.g.
will port tables update automatically etc.
· Which protocols are used in your local environment?
Will separate VLANs be required for each protocol? Is routing supported?
· What security features will be available and
of use?
Token Ring may be switched and benefits here are great. The delay waiting for the return of the token can be reduced and often the hop-count in Source Route Bridging SRB internets can be reduced. However token ring switches are more expensive than Ethernet switches due to a smaller market and technical challenges.
The technical challenges require more processor power:
· For token operations
· To process more of the frame in SRB environments
· SRB discovery frames can generate significant
traffic overheads
Larger buffers (than on equivalent Ethernet) are needed too:
· For store-and-forward due to waiting for token
· To cope with larger frame size (4500 byte frames)
Migration to Switched Token Ring
Replacing backbone token rings is the first step but port costs are still high and departmental and workgroup switches are still very expensive. Servers may benefit too from Switched Token Ring. Private Token Ring can boost throughput and full duplex operation is available with some products which allows for elimination of the token.
100 Mbit/sec Token Ring switching is now available.
Figure 6.13 Switched Token Ring Implementation
Overloaded FDDI rings suffering from network bottlenecks may benefit from switched architectures and vendors offer high performance switches with FDDI ports. The most likely deployment for switched FDDI is for Super-backbones and Back-end networks comprising of super-servers, supercomputers and mainframes, see fig 6.14.
Full duplex private FDDIs can achieve 200 Mbit/sec between dual-MAC A and B ports and also eliminate the need for the token. These switches are very very expensive currently.
Figure 6.14 A supercomputer
installation before upgrade
Figure 6.15 Switched FDDI Implementation
Switching allows an upgrade path that multiplies the existing bandwidth of the LAN and can allow full-duplex hub to station management. It can be included in an incremental fashion at the desktop, departmental and backbone level.
Switched Ethernet offers a simple and cost-effective method to upgrade existing Ethernet and Fast Ethernet LANs.
Token Ring and FDDI switching is available from some vendors.
(c)MMClements 2000 Back to top of page