IEEE 802.3 - Ethernet Eric Karlsson Thomas Karlsson Morgan Kejerhag
Today s Presentation Switching and bridging Auto negotiation Power-over-Ethernet Ethernet Performance Ethernet in WAN
Switching and Bridging Bridges Bridging methods Spanning Tree Protocol Switching hubs Switch based vlan
Bridges Connects network segments Works at OSI layer 2 Keeps a table of devices Forwarding Flooding
Bridges (cont.) Transparent bridges connects networks with the same protocol Translating bridges connects networks with different protocols
Bridges (cont.) Translating bridges connects networks with different protocols Ethernet Bridge Token-ring
Bridges (cont.) Design with translating bridges has a problem: All stations must use small frames since different protocols have different maximum size of frames
Bridges (cont.) Local bridge for LAN Remote bridge for LAN with connection to WAN
Bridging Methods Loops can not be allowed in bridged networks Ethernet 1 Ethernet 2 Bridge 1 Bridge 2 Bridge 3 Ethernet 3
Bridging Methods (cont.) Duplicate frames leads to Congestion Special reception at end-stations Physical loops are allowed Active loops are not allowed
Spanning Tree Protocol Solves the problem with loops Based on graph-theory Ensures a unique path from one station to another Puts unwanted bridges in standby mode Will activate standby mode bridges automatically if needed
Spanning Tree Protocol (cont.) Graph-theory Graphs constists of vertices connected by edges Path = a connected way from a node to another Tree = a loop free graph Edges can have costs =Vertex, Node =Edge
Spanning Tree Protocol (cont.) Example graph An edge between vertices A and B is noted (A,B) The cost of an edge is written next to the edge The cost of (B,D) is 3 A 1 4 C 8 6 B 3 D
Spanning Tree Protocol (cont.) A spanning tree is a tree that connects all nodes Not a spanning tree there is a loop! Not a spanning tree there is a disconnected node! A spanning tree!
Spanning Tree Protocol (cont.) There are usually many spanning trees for a graph
Spanning Tree Protocol (cont.) The first thing that happens when starting up a new bridged network is that a Root Bridge (RB) is chosen Every bridge B has a priority level P B2/P4 B4/P2 B1/P5
Spanning Tree Protocol (cont.) Every bridge starts by assuming that it is the RB and send messages to other bridges with its priority B2/P4 B4/P2 B1/P5
Spanning Tree Protocol (cont.) If a bridge receive a message from a bridge with higher priority it will stop assuming to be RB and stop sending new messages. B2/P4 B4/P2 B1/P5
Spanning Tree Protocol (cont.) Finally there is only one RB left, it is the real RB B2/P4 B4/P2 B1/P5
Spanning Tree Protocol (cont.) Cost is assigned to all bridge ports, for simplicity we put cost on edges here 10 B2 20 B4 B1 10 10 20 30
Spanning Tree Protocol (cont.) The RB starts sending a messages to connected bridges about the cost. The RB has cost 0 to itself 0 B2 20 B4 10 B1 10 10 20 0 30
Spanning Tree Protocol (cont.) The next bridges adds their costs and sends a message to their neighbours B2 20 B4 B1 10 10 10 10 10 20 20 20 30
Spanning Tree Protocol (cont.) This continues and each bridge determines which path will be the cheapest to the root 10 B2 20 B4 B1 10 10 20 30
Spanning Tree Protocol (cont.) After everything is done there is a spanning tree that is loop free B2 B4 B1
Spanning Tree Protocol (cont.) Simple example much more complex in reality Unused links can be used if another link goes down Disadvantage: no extra paths no possibility for load balance
Switching hubs More advanced bridge Multiple bridge operations at the same time Port switching Segment switching Enhanced performance
Switching hubs (cont.) Three main switching techniques Cross-Point Store-and-Forward Hybrid
Switching hubs (cont.) Cross-Point switches are also called cut-through or on-the-fly switches Examines the DA field of incoming frames and keeps a table of known addresses and what port they are associated with There is no error detection since only parts of frame is examined
Switching hubs (cont.) Low delay or latency 1 3 Switch engine 2 Look up table
Switching hubs (cont.) Store-and-forward switches stores full frame in buffer Can perform CRC check Can check length of frame If error the frame is discarded Some vendors have layer 2 filtering Also layer 3 filter, but is more like router
Switching hubs (cont.) Translating switches often support store-and-forward techniques since the frame needs to be stored anyway 1 2 Shared RAM Switch engine 3 Look up table 4
Switching hubs (cont.) Hybrid switches are both cut-through and store-and-forward Starts in cut-through mode and monitors error rate If error rate is high the switch will change to store-and-forward mode Provides minimum delay when error rate is low Discards invalid frames when error rate is high
Switch based vlan Virtual LAN is a LAN inside the LAN itself Requires special equipment Layer 2 switches Layer 3 routers Implicit vs Explicit tagging
Switch based vlan (cont.) In explicit tagging is the vlan identification sent as an extra field in each frame IEEE standard 802.1Q Creates problems for frames of already maximum length
Switch based vlan (cont.) 802.1Q Frame 16 bit Tag Protocol Identifier (TPI) 3 bit priority (802.1p) 1 bit Canonical Format Identifier (CFI) 12 bit vlan Identification (VID) Preamble SFD DA SA 802.1Q TagType Tag Control Data FCS TPI Priority CFI VID
Switch based vlan (cont.) Implicit tagging for example: All stations connected to port 1, 2 and 4 belongs to vlan X Vendors have different standards => Can t connect equipment from different vendors
Switch based vlan (cont.) Port based vlans The first method A certain port belongs to a certain vlan 0, 1, 5, 6 to vlan1 2, 3, 4, 7 to vlan2 0 1 2 3 Switch matrix Not flexible 4 5 6 7
Switch based vlan (cont.) MAC based vlans A certain device (determined by MAC address) belongs to a certain vlan Full flexibility 0 1 2 3 Creates problems for higher level protocols (can be handled) Switch matrix 4 5 6 7
Auto negotiation What is auto negotiation and why is it a nice feature. How does it work? More benefits. Auto negotiation in gigabit networks.
What is Auto negotiation? Auto negotiation is the process when two devices automatically configure themselves for the highest possible transfer mode. Auto negotiation became useful when we started to get networks mixed with 10Mbit/s and 100Mbit/s devices.
Auto negotiation an example Imagine the following: The workstations support 100Mbit/s but the switch is an old 10Mbit/s device all the cards have then been set to 10Mbit/s in order to talk to each other. Now imagine that the switch is upgraded to a 100Mbit/s device, things will no longer work until all the workstations Ethernet adapters have been manually configured to 100Mbit/s.
Auto negotiation an example (cont.) All this work for the system administrator can be solved using auto negotiation capable devices. In the above example all the Ethernet adapters would set themselves to operate at 100Mbit/s automatically as soon as the switch was replaced, given that all adapters support auto negotiation.
How does it work? Auto negotiation relies upon a modification in the link integrity test used by 10BaseT. Instead of sending a link test pulse a series of fast test pulses are sent. These fast test pulses form a 16bit link code word.
The Fast Link Pulse (FLP) NLP FLP 2 ms 8-24 ms Typically 16 ms
How are the FLP s interpreted?... 1 0 1 100 ns ~60 us Clock pulses (17) data pulses (0-16)
How does it work? (cont.) S0 S1 S2 S3 S4 A0 A1 A2 A3 A4 A5 A6 A7 RF ACK NP The first 5 bits are used to determine what MAC protocol to use. A0 A7 is used to indicate what operation modes that are supported. A0 = 10BASE-T A1 = 10BASE-T Full Duplex A2 = 100BASE-TX A3 = 100BASE-TX Full Duplex A4 = 100BASE-T4 A5 = Pause
How does it work? (cont.) S0 S1 S2 S3 S4 A0 A1 A2 A3 A4 A5 A6 A7 RF ACK NP The last three bits are control bits. RF = Remote Fault Indicates an error in communication. ACK = Acknowledge Used to tell the other device that it s last code word was properly received. NP = Next Page If one the device is supporting the exchange of additional information.
How is the mode negotiated After both devices have received information about the other one. The transfer mode is determined by a priority table. 1. 100BASE-TX Full Duplex 2. 100BASE-T4 3. 100BASE-TX 4. 10BASE-T Full Duplex 5. 10BASE-T
More benefits Does not require any intervention from a user or network administrator. Backward compatibility. Network protection Easier upgrade
Problems with different speed What happens if a switch has one 10Mbit/s device and one 100Mbit/s device connected to it? Three different approaches Do nothing. Send a JAM signal Send pause frames to the sending unit (only in full duplex)
Auto negotiation in gigabit Ethernet Only supported on copper media. Additional pages need to be sent. Extended priority table.
The additional link code words First one normal LCW is sent. Then a message page M0 M1 M2 M3 M4 M5 M6 M7 M8 M9 M10 T Ack2 MP Ack NP The MP bit indicates that it is a message page. There is only a limited number of messages. For example 8 00010000000 indicate 1000BaseT.
The additional link code words (cont.) After the message page the technology is further described in two unformatted pages. U0 U1 U2 U3 U4 U5 U6 U7 U8 U9 U10 T Ack2 MP Ack NP The first unformatted page is interpreted as follows: U10:U5 reserved U4 1000BaseT half duplex U3 1000BaseT full duplex U2 Port type (1 = multiport, 0 = singelport) U1 1 = master, 0 = slave (ignored if U0 = 0) U0 Manually configure master/slave (0 = automatic, 1 = manual) The last page contains the master-slave seed value
Priority table The extended priority table looks like this: 1. 1000BaseT Full duplex 2. 1000BaseT Half duplex 3. 100BaseT2 Full duplex 4. 100BaseTX Full duplex 5. 100BaseT2 Half duplex 6. 100BaseT4 7. 100BaseTX Half duplex 8. 10BaseT Full duplex 9. 10BaseT Half duplex
Power over Ethernet (PoE) Power over Ethernet is all about distribute electrical power over TP cable. PoE can be used with existing hardware. PoE greatly increases flexibility
Power distribution Typically 48V DC (min 36, max 57) Two valid approaches Over the data pairs Over the unused pairs A Powered device is required to support both methods while a power source is only required to support one of them.
Power distribution (cont.) Can be distributed by PoE switches which have built in power supply Support 10BaseT, 100BaseTX and 1000BaseT Can be distributed by power injectors placed between the switch and the end station. Must distribute the power over the unused pairs. Only support 10BaseT and 100BaseTX
The powered device The PD must present a valid detection signature when it accepts power but are not receiving power. The PD must not present a valid detection signature if it won t accept power. When powered it must present a invalid detection signature on the set of pairs from witch it isn t drawing power from. Must use a minimum current.
The powered device (cont.) A powered device can have different classes. In addition to a valid detection signal a PD can provide one and only one classification signature. Class 0 1 2 3 4 Usage Default Optional Optional Optional Not allowed Range of maximum power 0.44 to 12.95 Watts 0.44 to 3.84 Watts 3.84 to 6.49 Watts 6.49 to 12.95 Watts Reserved
Backward compatibility Non PoE devices must work without the risk of getting fried. A detection process is used. A low, current controlled, voltage is send. The PSE is then able to detect if the device has a 25k ohm resistor. If it has all 48 volts are applied.
Limitations A powered device is allowed to consume 12.95 W. Sufficient for many devices, such as IP-phones and Wireless access points However it s not enough to use your laptop without battery or printers and other high consuming devices.
Limitations (cont.) Requires additional power to your wiring closets If a UPS is used to maintain the network in case of a power outage it might have to be upgraded What happens if the UPS cannot support all the PD s Every PD will shutdown The PSE should be able to selectively shut down PD s
The main benefits The biggest benefit is when you install permanent low consuming equipment. Greater flexibility when installing devices. No need to get an electrician to rewire your power cables It can potentially save a lot of money.
Ethernet performance
Myths and facts Ethernet saturates at 37% load Simplified model, 256 stations Real-life Ethernet above 95%
Performance graph Delay (s) 100 10 1 N 1 5 2 1 5 0,1 0,01 0,001 0,0001 0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1
Light load (0%-50%) Very low response times 0.001 second or lower No issues Real-time applications work well Bulk applications work well Network is healthy
Moderate to heavy load (50%-80%) Low response times 0.01 second to 0.1 second Some issues Real-time applications might experience some loss Bulk application work well Network is on the brink of capacity
Very heavy load (80%-100%) Long response times Up to 1 second or more Major issues Real-time applications impossible Bulk applications experience performance loss Network is badly undersized
Ethernet in WAN
Overview Ethernet leader in LAN Ethernet is cheap Ethernet is everywhere Why not use it in WAN as well?
OAM Operation, Administration, Management Tools for diagnostics and fault detection
Service differentiation Guarantee bandwidth Divide services and customers Police traffic QoS (Quality of Service)
SONET and ATM SONET High availability Quick failover (50 ms) Ring structure Reliable Many services OAM
SONET and ATM (cont.) ATM Service differentiation Circuit switched Guaranteed bandwidth OAM
10 Gigabit Ethernet 10 Gigabit Ethernet SONET compatible Service differentiation 802.1p, 802.1q Reliable and scalable OAM (partly through SNMP) Cheap Ring structure (multiple spanning tree) Rate limiting (vendor feature)
Ethernet chances + Cheap + Compatible + Features - New market segment - 10G segment already populated - SONET OC-768 @ 40 Gbps
That s all folks!