ch06-switching

ch06-switching - Chapter 6 Switching Master Supcom, Fall...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Chapter 6 Switching Master Supcom, Fall 2006 What is it all about s How do we move traffic from one part of the network to How another? another? Connect end-systems to switches, and switches to Connect each other each Data arriving to an input port of a switch have to be Data moved to one or more of the output ports moved Types of switching elements: s s s Telephone switches: switch samples Datagram routers: switch datagrams ATM switches: switch ATM cells Classification Packet vs. circuit switches packets have headers and samples don’t packets Connectionless vs. connection oriented Connectionless connection oriented switches need a call setup setup is handled in control plane by switch controller setup control connectionless switches deal with self-contained datagrams Other switching functions s s Participate in routing algorithms to build routing tables Resolve contention for outputs: for scheduling purposes Admission control: to guarantee resources to certain streams Buffering: to much input rate to the service Perform multiplexing and demultiplexing s s s Requirements s s Capacity of switch is the maximum rate at which it can Capacity move information, assuming all data paths are simultaneously active simultaneously Primary goal: maximize capacity Primary maximize Constraints : subject to cost and reliability requirements s Circuit switch must reject call if it cannot find a path for Circuit samples from input to output samples subgoal: minimize call blocking subgoal: minimize s Packet switch must reject a packet if it cannot find a Packet buffer to store it awaiting access to output trunk buffer subgoal: minimize packet loss subgoal: minimize s Don’t reorder packets Summary s Multiplexing Frequency division multiplexing Synchronous Time division multiplexing Statistical time-division multiplexing s s s Circuit switching Message Switching Packet switching Switch generations Switch fabrics Buffer placement I. Multiplexing Three types of multiplexing techniques are considered. Three The first frequency division multiplexing (FDM) is the most widespread and is familiar to anyone. The second is a particular case of time division multiplexing (TDM) often known as synchronous TDM. this is a commonly used for multiplexing digital voice streams. The third type seeks to improve on the efficiency of synchronous TDM by adding complexity to the multiplexer. Frequency-Division Multiplexing Frequency-division multiplexing is possible when the useful bandwidth of the medium exceeds the required bandwidth of signals to be transmitted. A number of signals can be carried simultaneously if each signal is modulated onto a different carrier frequency, and the carrier frequencies are sufficiently separated that the bandwidths of the signal do not overlap. The composite signal transmitted across the medium is analog. analog. However, the input signals may be either digital or However, analog. Synchronous Time-division Multiplexing Synchronous time-division multiplexing is possible when Synchronous the achievable data rate of the medium exceeds the data rate of digital signals to be transmitted. Multiple digital signals can be carried on a single transmission path by interleaving portions of each signal in time. The interleaving can be at the bit level or in blocks of bytes or larger quantities. Time slot 12 Frame n 12 Frame n May be empty Generic depiction of a synchronous TDM A number of signals [mi(t), i =1,n] are to be multiplexed onto the same medium. Each buffer is typically one bit or one character in length. The buffers are scanned sequentially to form a composite digital data stream m(t). the scan operation is sufficiently m(t). rapid so that each buffer is emptied before more data can arrive. The data rate of m(t) must be at least equal to the m(t) sum of the mi(t), At the receiver, the interleaved data are demultiplexed and routed to the appropriate destination buffer. and TDM Link Control s Flow control error control are provided using concepts such as framing, stuffing, and synchronizing clocks. framing, If each source has a separate clock, any variation among clocks If could cause loss of synchronization. Flag or sync characters can be used. be In some cases the data rates of the input data streams are not related by a simple rational number. An extra capacity is used by stuffing extra bits or pulses into each incoming signal until its rate is raised to that of locally generated clock signal. rate The stuffed pulses are inserted in the multiplexer frame format so The that they may be identified and removed at the demultiplexer that s s s Example s Consider that there are 11 sources to be multiplexed on a single link: link Source 1. Analog, 2-kHz bandwidth Source 2. Analog, 4-kHz bandwidth Source Source 3. Analog, 2-kHz bandwidth Source 4-11. Digital, 7200 bps synchronous Source s s s The analog sources are converted to digital by sampling. The required sampling rate is 4000 for sources 1 and 3, and 8000 samples for sampling source 2. Assume that 4 bits are used for each analog sample. The he analog are multiplexed first. analog For the digital sources, pulse stuffing is used to raise each source to For a rate of 8kbpsfor an aggregate data rate of 64 kbps. rate Thus a total of 16 bits is generated at a rateof 4000 per second Sourse 1 2 KHz, analog 16 KHz 4 bit A/D 16 KHz 16 – bit buffer Sourse 2 4 KHz, analog ƒ ƒ = 4khz Sourse 3 2KHz, analog Sourse 4 7.2 KHz, digital Pulse stuffing 8kdps 2 – bit buffer TDM signal 128 kdps Scan Sourse 11 7.2 KHz, digital Pulse stuffing 8 kbps 2 – bit buffer Statistical time-division multiplexing s As with synchronous TDM, the statistical multiplexer has a As number of I/O lines on one side and a higher speed multiplexed number line on the other. Each I/O line has a buffer associated with it. line I/O In the case of statistical multiplexer, there are n I/O lines, but only k, k< n, time slots available on the TDM frame. For input, only the function of the multiplexer is to scan the input buffers, collecting data until frame is filled, and then send the frame. collecting On output, the multiplexer receives a frame and distributes the On slots of data to the appropriate output buffers. to FLAG Address Control TDM subframes FCS Flag s Overall frame Performance s s s s s Define for a statistical time division multiplexer: Define for division N = number of input sources, R data rate of each source, bps. M = effective capacity of multiplexed line, bps. effective α = mean fraction of time each source is transmitting, 0< α < 1. mean 0< K = M/NR = ratio of multiplexed line capacity to total maximum input. maximum s Some results can be obtained by viewing the multiplexer as a single-server queue single-server Delay incurred by a customer (time spent waiting in the queue + time to be Delay served) served) Speed of the multiplexed line Average queue length s s s II. Circuit switching s s s Moving 8-bit samples from an input port to an output port Samples have no headers Destination of samples depends on time at which it arrives at the Destination time switch Multiplexers represent simpler mechanisms than switches: s Most trunks time division multiplex voice samples At a central office, trunk is demultiplexed and distributed to active circuits Synchronous multiplexer (N input lines, Output runs N times as fast as Synchronous input) input) Circuit multiplexing 2 Communication via circuit switching involves three phases : Communication Circuit establishment: An end-to-end circuit must be establish. Based on Circuit routing information and measures of availability and perhaps cost links are selected to form a path from the sender to the receiver Data transfer: Signals can now be transmitted from the sender through the Data network to receiver. Circuit disconnect: After some period of data transfer, the connection is Circuit terminated, usually by the action of one of the two stations. s Constraints. Both stations must be available at the same time for the Constraints. data exchange. Resources must be available and dedicated through the network between the two stations the Call blocking s s Problem: Cannot find a path from input to output Internal blocking slot in output frame exists, but no path s Output blocking 1. no slot in output frame is available s Output blocking is reduced in transit switches Output switches need to put a sample in one of several slots going to the need several desired next hop III. Message Switching s s With message switching, it is not necessary to establish a With dedicated path between two stations. Rather, if a station wishes dedicated to send message it appends a destination address to the message to (e.g. electronic mail, files, transactions,..). The message is then (e.g. passed through the network from node to node. At each node the passed entire message is received, stored briefly, and then transmitted to entire the next node. the In a circuit-switching network, each node is an electronic or perhaps electromechanical switching device which transmits bits perhaps as fast as it receives them. A message-switching node is typically a general purpose minicomputer, with sufficient storage to buffer messages as they come in. A message is delayed at each node for messages the time required to receive all bits of the message. to receive Advantages of message-switching over CS s s s s Line efficiency is greater, since a single node-to-node channel can be shared by many messages over time. For the same traffic volume, less total transmission capacity is needed. traffic Because messages over time may be temporarily stored at any Because point en route. Simultaneous availability of sender and receiver are not required. The network can store the message pending the availability of the receiver. availability When traffic becomes heavy on a circuit-switched network, some calls are blocked. On a message-switched network, messages are still accepted, but delivery delay increases. A message-switching system can send one message to many destinations. destinations. More advantages s s s s Message priorities can be established. If a node has a Message number of number of messages queued for transmission, it number can transmit the higher-priority messages first. can Error control and recovery procedures on a message basis Error can built into the network. messages may be numbered and can a copy filed for later retrieval in case the original fails. copy retrieval A message-switching network can carry out speed message-switching conversion. It can also easily convert format. conversion. Messages sent to inoperative may be intercepted and either stored or rerouted to other terminals. stored IV. Packet switching s Packet switching represents an attempt to combine the advantages of Packet message and circuit switching while minimizing the disadvantages. Packet switching is like message switching. The difference is that the length Packet of units of data that may be presented to the network is limited in a packet switched network. Messages above the maximum length must be divided into smaller units and sent out one at a time. Another difference from message switching is that packets are typically not Another filed. The network will handle the stream of packets based on two approaches: approaches: Datagram approach. Each packet is treated independently, just as each message is treated independently. Virtual circuit approach. A logical connection is established before any Virtual packets are sent. Each packet now contains a virtual circuit identifier as well as data. s s s s Packet size/transmission time s There is a significant relationship between packet size and transmission time. Example. Assume that there is a virtual circuit from station X through nodes a and b to station Y. The message to be sent contains 30 octets, and each packet contains 3 octets of control contains information. If the entire message is sent as a single message of 33octets, then the total transmission time at the nodes is 99 octet-times. If the message is broken up into 2 packets, then the total time is 72 octetIf time times s s Effect of packet size on transmission time. Header (a) 1-packet message Date 1 (b) 2-packet message (c) 5-packet (d) 10-packet 1 2 1 2 3 4 5 message 1 2 3 1 2 3 4 5 6 7 8 9 10 message Data 1 2 3 4 5 6 7 8 9 10 Time Header Date Date 2 1 3 4 1 2 3 4 5 4 5 6 7 Data Date Date 2 1 5 8 9 10 Header Date 2 Data X a b Y X a b Y X a b Y X a b Y Blocking in packet switches s s s s Can have both internal and output blocking Internal: no path to output Internal: Output: trunk unavailable Output: Unlike a circuit switch, cannot predict if packets will Unlike block If packet is blocked, must either buffer or drop it s Dealing with blocking s Over provisioning: Over internal links much faster than inputs s Buffers Buffers at input or output) s Backpressure: if switch fabric doesn’t have buffers, Backpressure: prevent packet from entering until path is available prevent Parallel switch fabrics Parallel s increase effective switching capacity Three generations of packet switches s s Different trade-offs between cost and performance Represent evolution in switching capacity, rather than in Represent technology technology With same technology, a later generation switch achieves With greater capacity, but at greater cost greater s All three generations are represented in current All products products First generation switch s s Most Ethernet switches and cheap packet routers Bottleneck can be CPU, host-adaptor or I/O bus, Bottleneck depending depending Second generation switch s s s Port mapping intelligence in line cards ATM switch guarantees hit in lookup cache Ipsilon IP switching Ipsilon IP 1. 2. 3. assume underlying ATM network by default, assemble packets iif detect a flow, ask upstream to send on a particular VCI, and install f entry in port mapper => implicit signaling entry Third generation switches s s Bottleneck in second generation switch is the bus Third generation switch provides parallel paths (fabric) Switch fabrics s Transfer data from input to output, ignoring Transfer scheduling and buffering scheduling Usually consist of links and switching elements Usually switching s Crossbar s Simplest switch fabric x think of it as 2N buses in parallel s Used here for packet routing: crosspoint is left open Used packet long enough to transfer a packet from an input to an output output For fixed-size packets and known arrival pattern, can For compute schedule in advance compute Otherwise, need to compute a schedule on-the-fly Otherwise, (what does the schedule depend on?) (what s s Buffered crossbar s What happens if packets at two inputs both want to go What to same output? to Can defer one at an input buffer Or, buffer crosspoints s s Broadcast s s s s Packets are tagged with output port number Each output matches tags Need to match N addresses in parallel at each output Useful only for small switches, or as a stage in a large Useful switch switch Blocking s Can avoid by checking if path is available before sending Can packet packet three-phase scheme: send requests, inform winners, three-phase send packets send Or, use several fabrics in parallel s intentionally misroute and tag one of a colliding pair divert tagged packets to a second fabric switch divert can reorder packets output buffers have to run k times faster than input Effect of packet size on switching fabrics s A major motivation for small fixed packet size in ATM is major ease of building large parallel fabrics ease In general, smaller size => more per-packet overhead, In but more preemption points/sec but s At high speeds, overhead dominates! s Fixed size packets helps build synchronous switch But we could fragment at entry and reassemble at exit Or build an asynchronous fabric Thus, variable size doesn’t hurt too much s Maybe Internet routers can be almost as cost-effective Maybe as ATM switches as Buffering All packet switches need buffers to match input rate to All service rate service or cause heavy packet loses Where should we place buffers: input in the fabric output shared Input buffering (input queuing) s No speedup in buffers or trunks (unlike output queued No switch) switch) Needs arbiter Problem: head of line blocking Problem: head s s with randomly distributed packets, utilization at most 58.6% worse with hot spots worse hot Dealing with HOL blocking s s Per-output queues at inputs Arbiter must choose one of the input ports for each Arbiter output port output Parallel Iterated Matching inputs tell arbiter which outputs they are interested in output selects one of the inputs some inputs may get more than one grant, others some grant others may get none may if >1 grant, input picks one at random, and tells if output output losing inputs and outputs try again s Output queuing s s s Don’t suffer from head-of-line blocking But output buffers need to run much faster than trunk speed Can reduce some of the cost by using the knockout principle Can knockout unlikely that all N inputs will have packets for the same output drop extra packets, fairly distributing losses among inputs Shared memory s s Route only the header to output port Bottleneck is time taken to read and write multiported Bottleneck memory memory Doesn’t scale to large switches But can form an element in a multistage switch s s ...
View Full Document

This note was uploaded on 03/31/2010 for the course SUPCOM 1 taught by Professor Boudriga during the Spring '10 term at Hibbing CC.

Ask a homework question - tutors are online