HP 180 Degree Turn User Manual

The Ethernet Evolution

The 180 Degree Turn

2005/03/11(C) Herbert Haas

“Use common sense in routing cable. Avoid wrapping coax around sources

of strong electric or magnetic fields.

Do not wrap the cable around

flourescent light ballasts or

cyclotrons, for example.”

thernet Headstart Product, Information and Installation Guide,

Bell Technologies, pg. 11

History: Initial Idea

 Shared media  CSMA/CD as access algorithm  COAX Cables  Half duplex communication  Low latency  No networking nodes

(except repeaters)

 One collision domain and also one broadcast domain

10 Mbit/s shared

by 5 hosts  2

Mbit/s each !!!

2005/03/11

The initial idea of Ethernet was completely different than what is used today under the term "Ethernet". The original new concept of Ethernet was the use of a shared media and an Aloha based access algorithm, called Carrier Sense Multiple Access with Collision Detection (CSMA/CD). Coaxial cables were used as shared medium, allowing a simple coupling of station to bus-like topology.

Coax-cables were used in baseband mode, thus allowing only unicast transmissions. Therefore, CSMA/CD was used to let Ethernet operate under the events of frequent collisions.

Another important point: No intermediate network devices should be used in order to keep latency as small as possible. Soon repeaters were invented to be the only exception for a while.

An Ethernet segment is a coax cable, probably extended by repeaters. The segment constitutes one collision domain (only one station may send at the same time) and one broadcast domain (any station receives the current frame sent). Therefore, the total bandwidth is shared by the number of devices attached to the segment. For example 10 devices attached means that each device can send 1 Mbit/s of data on average.

Ethernet technologies at that time (1975-80s): 10Base2 and 10Base5

History: Multiport Repeaters

 Demand for structured cabling (voice-grade

twisted-pair)

 10BaseT (Cat3, Cat4, ...)

 Multiport repeater ("Hub") created  Still one collision domain

("CSMA/CD in a box")

Later, Ethernet devices supporting structured cabling were created in order to reuse the voice-grade twisted-pair cables already installed in buildings. 10BaseT had been specified to support Cat3 cables (voice grade) or better, for example Cat4 (and today Cat5, Cat6, and Cat7).

Hub devices were necessary to interconnect several stations. These hub devices were basically multi-port repeaters, simulating the half-duplex coax-cable, which is known as "CSMA/CD in a box". Logically, nothing has changed, we have still one single collision and broadcast domain.

Note that the Ethernet topology became star-shaped.

2005/03/11

History: Bridges

 Store and forwarding according destination MAC

address

 Separated collision domains  Improved network performance  Still one broadcast domain

Three collision

domains in this

example !

2005/03/11

Bridges were invented for performance reasons. It seemed to be impractical that each additional station reduces the average per-station bandwidth by 1/n. On the other hand the benefit of sharing a medium for communication should be still maintained (which was expressed by Metcalfe's law).

Bridges are store and forwarding devices (introducing significant delay) that can filter traffic based on the destination MAC addresses to avoid unnecessary flooding of frames to certain segments. Thus, bridges segment the LAN into several collision domains. Broadcasts are still forwarded to allow layer 3 connectivity (ARP etc), so the bridged network is still a single broadcast domain.

History: Switches

 Switch = Multiport Bridges with HW acceleration  Full duplex  Collision-free Ethernet  No CSMA/CD

necessary anymore

 Different data rates at the same time supported

 Autonegotiation

 VLAN splits LAN into several broadcast domains

Collision-free

plug & play

scalable Ethernet !

0 Mbit/s

2005/03/11

100 Mbit/s

1000 Mbit/s

00 Mbit/s

Several vendors built advanced bridges, which are partly or fully implemented in hardware. The introduced latency could be dramatically lowered and furthermore other features were introduced, for example full duplex communication on twisted pair cables, different frame rates on different ports, special forwarding techniques (e,g, cut through or fragment free), Content Addressable Memory (CAM) tables, and much more. Of course marketing rules demand for another designation for this machine: the switch was born.

Suddenly, a collision free plug and play Ethernet was available. Simply use twisted pair cabling only and enable autonegotiation to automatically determine the line speed on each port (of course manual configurations would also do). This way, switched Ethernet become very scalable.

Furthermore, Virtual LANs (VLANs) were invented to split the LAN into several broadcast domains. VLANs improve security, utilization, and allows for logical borders between workgroups.

Today

 No collisions  no distance limitations !  Gigabit Ethernet becomes WAN

technology !

 Over 100 km link span already

 Combine several links to "Etherchannels"

 Acts as single link from the spanning-tree view

• Cisco: Port Aggregation Protocol (PAgP)

• IEEE 802.1ad: Link Aggregation Control Protocol

(LACP)

1 Gbit/s or even 10 Gbit/s long reach connection !!!

Today, Gigabit and even 10 Gigabit Ethernet is available. Only twisted pair and more and more fiber cables are used between switches, allowing full duplex collision-free connections. Since collisions cannot occur anymore, there is no need for a collision window anymore! From this it follows, that there is virtually no distance limit between each two Ethernet devices.

Recent experiments demonstrated the interconnection of two Ethernet Switches over a span of more than 100 km! Thus Ethernet became a WAN technology! Today, many carriers use Ethernet instead of ATM/SONET/SDH or other rather expensive technologies. GE and 10GE is relatively cheap and much simpler to deploy. Furthermore it easily integrates into existing low-rate Ethernet environments, allowing a homogeneous interconnection between multiple Ethernet LAN sites. Basically, the deployment is plug and play.

If the link speed is still too slow, so-called "Etherchannels" can be configured between each two switches by combining several ports to one logical connection. Note that it is not possible to deploy parallel connections between two switches without an Etherchannel configuration because the Spanning Tree Protocol (STP) would cut off all redundant links.

2005/03/11

Depending on the vendor, up to eight ports can be combined to constitute one "Etherchannel".

What About Gigabit Hubs?

 Would limit network diameter to 20-

25 meters (Gigabit Ethernet)

 Solutions

 Frame Bursting  Carrier Extension

 No GE-Hubs available on the market

today  forget it!

 No CSMA/CD defined for 10GE (!)

Remember: Hubs simulate a half-duplex coaxial cable inside, hence limiting the total network diameter. For Gigabit Ethernet this limitation would be about 25 meters, which is rather impracticable for professional usage. Although some countermeasures had been specified in the standard, such as frame bursting and carrier extension, no vendor developed an GE hub as for today. Thus: Forget GE Hubs!

The 10 GE specification does neither consider copper connections nor hubs. 10 GE can only run over fiber.

At this point please remember the initial idea in the mid 1970s: Bus, CSMA/CD, short distances, no network nodes.

Today: Structured cabling (point-to-point or star), never CSMA/CD, WAN capabilities, sophisticated switching devices in between.

2005/03/11

+ 16 hidden pages