High Speed LANs

Contents

• Faster nets ... why?
• Theory
• Buying Equipment
• Migration Concerns
• Example Designs
• Foretelling the Future
• Sources and For More Info

Traffic Patterns: Cisco's 80/20 Rule

Virtually all of Cisco's documents discuss a rapid change in traffic patterns for many local networks. Previously most of the traffic remained local, however now Cisco claims that this has changed and now typically only 20% remains local and the remainder goes elsewhere. Many areas on campus still have in-house servers and the bulk of their traffic remains local. Discussions about traffic flows usually give rise to considerations of ...

Congestion, Utilization/Capacity

One of the reasons often provided for wanting to implement higher speed networking is a perception that it will take care of all the congestion and make everything go faster. Not necessarily so!! One of the most important things you can do when planning network upgrades is to determine just where your traffic is flowing.

There are a myriad of monitoring packages that can help you gauge the utilization of your various segments and to determine if there is congestion. Some of these are:

• Use EtherPeek, Lanalyzer, etc., to view statistics
• Networking has Sniffers that monitor traffic
• View the data from Ethermeters on your subnet via Meterware Reports
• View router stats via Meterware Reports
• Check stats from various interfaces to view dropped packets, re-sends, defers, errors, etc.
• Many devices now collect statistics: accessible via Telnet and SNMP/RMON

Note that some of these are not available for Fast Ethernet, e.g., we currently have only 10 Mbps Sniffers.

Shared vs. Switched

This will be discussed in more detail a little later ... in shared environments every node hears every conversation between devices; switches bracket off private conversations while allowing the packets meant for everyone to be distributed accordingly. This is mentioned in this "Why?" section only to raise an important question: which is more helpful ... having switched 10 or shared 100?

New Features/Price Reduction

A very good reason for considering alternatives to shared 10 equipment is that new integrated circuits have come out that have allowed manufacturers to make switched 10 and switched 10/100 much more economically and with more sophistocated features.

It's Inevitable: Needed or Not!

The faster technologies eventually will become the standard. And there's a corollary...

Politics of New Technologies

THEY have Fast Ethernet, so WE should too!

Carrier Sense Multiple Access/Collision Detect (CSMA/CD)

Just as in all other ethernet forms, FastEthernet (IEEE 802.3u) is based upon the ability for a node to start transmitting when it detect silence ... as well as determine when a different node talks at the same time (a collision).

Frame Format

The packet fields are identical in structure to those in 10BaseT

Bit Rates and Slot Times



A Helpful Table ...

Speed	Bit Time	Slot Time
10 Mbps	100 ns	51.2 µs
100 Mbps	10 ns	5.12 µs
1000 Mbps	1 ns	0.512 µs

A Digression: Quick Physical Layer Overview



Signal Characteristics:

• 100FX: 125 MHz (4B5B encoding of 100 Mbps input, converted to light pulses)
• 100TX: 100 MHz (scrambled 100 Mbps input, converted to three level signal: MLT3)

Both scrambling and 4B5B encoding assure adequate transitions in the bit stream to allow clock synchronization, as well as taking care of any DC bias issues.



Media: 100TX; 100FX; 100T4; MII

Note that both copper and fiber media have Distance-Bandwidth specifications:

Frequency	MM Fiber	Cat 5
250 MHz	2 km	100m
500 MHz	1 km	50m
1000 MHz	500m	25m

Size of Collision Domain:

As discussed in the sections above, we know the Slot Time for Fast Ethernet to be 5120 ns. This information can now be used to determine configuration specifications in one of two ways ... bit time calculations and distance measures.

Network Interface Cards (NICs)

1000 ns set aside for processing (100 bit times)

Class I and II Repeaters

There are two types of repeaters, the distinguishing factor is the maximum amount of time allotted for signal processing:

• Class I: 1400 ns (140 bit times)
• Class II: 920 ns (92 bit times)

Class I repeaters typically use the additional time for translating signals from one media to another, e.g., 100BaseTX to 100BaseFX.

Category 5 UTP and MM Fiber

The balance of the time is available for transit through the connections. But just how long can these connections be? It depends on how fast the signals propagate through the media. This is expressed by the Nominal Velocity of Propagation (NVP) and is expressed as a percentage of the speed of light (3 x 10^8 m/s). If you work it all out, light in a vacuum takes 3.33 ns to travel a meter, or a round-trip time of 6.67 ns/m. Using a NVP of 66.67% yields a nifty conversion factor of 10ns/m (and SU's Cat5 specs are actually higher ... 72%) so that we can change over the bit time numbers to distances ... as shown in the following section.

Simplified Design Rules

Much simpler to do:

• 412 m maxmimum distance allowed
• 140 m deducted for each Class I repeater
• 92m deducted for each Class II repeater

Stretching the Rules?

Note that SU's NVP is higher ... so that you actually have a comfortable "cushion"!

Hublets and Shared 100: DANGER!

The above measures are for shared 100Base connections. Say you have Class I repeater in the nearby wiring closet ... if someone connects a hublet to a TSO being served by that closet, you immediately go out-of-spec! Arrrgh! There are only two ways around this problem: to ban such devices from your building, or to use a Fast switch instead of shared repeater. Since switched ports are bridged (and thus separate one collision domain from another) the hublet dangling off the TSO is the ONLY Class I repeater in it's domain.

Overcoming Distance Limitations: Switched Full-Duplex

Switched Full-Duplex connections (where sending and receiving is occurring simultaneously and independently) can extend the distance limitations, e.g., 100BaseFX connections can go over 2 km! This would most likely be used for runs to the gateway or to other buildings.

Switching Issues:

• Modes:
  • Cut Through - forwarded as soon as source/destination received
  • Store & Forward - forwarded only after entire packet received (default when converting between 10 and 100)
  • In-between - forwarded after the 64 byte boundary is met

• Buffers: whether pooled or dedicated to specific ports, the switch stuffs the packets here until they can be sent. Once the buffers are full, incoming packets are ignored.
• Virtual LANS (VLANs): ports can be grouped together so as to see only the broadcast traffic destined for that group. VLANs from various switches can be interconnected via multiple links, or via a single link (or trunk) when using Cisco's Interswitch Link (ISL). ISL works across routers as well.
• Multicast traffic control: Most switches have methods to direct IP multicast to those specific ports which have subscribed via IGMP.
• Layer 3 Switching: can bypass router for installations with multiple subnets.

Types:

• Chassis w/ modules
• Stacked.
• Stand-alone
• Hublets

Feature-Set:

• Shared vs. Switched
• Media Variety
• Backplane and other Bottlenecks
• Switching modes
• VLAN/ISL/CGMP/Layer 3

Example #1: ChemEng's Shared 100FX Backbone

The Chem Engineering distribution device uses a shared 100FX module, which worked fine for short intra building backbone connections where no user could plug in devices to pull the net out-of-spec.

---

Chassis Design w/ Upgradeable Modules

This was proposed design was actually implemented.

---

Stackable Approach

This design was not implemented - instead low cost modular switches arranged in star form were employed rather than the Asante stackable switches.

---

Switch w/ Multiple VLANs and Subnets

---

• Gigabit Ethernet: Distance Limitations Revisited
• ATM/SONET between locations

• Spurgeon's Ethernet Site
• Intel's Fast Ethernet Deployment Guide
• Asante's Website
• Bay Networks
• Cisco's Website
• Gigabit Ethernet Alliance
• 3Com's Website