ATM - notes


Connection - oriented -> virtual circuit
-> data stream from origin to destination over same path

VCI -> virtual channel identified -> unique for each virtual circuit

 - overhead -> number of bits needed to identify VC - generally
 much less than full source/destination address of datagram

- cells reach destination in order sent from source so no sequence number, no
 buffering at destination for re-ordering (do have seq. number in application
 level if need to detect cell loss)

- switches identify different connections by VCI - can use for many different
 things:

- admission control -> refusing entry if not enough NW resources available
- congestion control -> limiting amount of traffic from connection
- resource allocation (negotiating BW & buffer allocation)
- policing - monitoring burstiness & average rate of traffic in connection

Disadvantages of connection-oriented

 - overhead of connection set-up when only few cells to be transferred -> more 
efficient to use datagram service

 - link or node failure terminates VC while for datagram only few packets lost


5 categories of service ATM can provide:

- CBR -> constant bit rate
- VBR-RT -> variable bit rate - real time
- VBR-NRT -> VBR - non-RT
- ABR -> available bit rate
- UBR -> unspecified bit rate

Parameters of traffic defined by GCRA  (generalized cell rate algorithm) has:

PCR - peak cell rate
SCR - sustained cell rate
CDVT - cell delay variation tolerance - how much variance from periodicity of
          traffic allowed
BT - burst tolerance
MCR - minimum cell rate 


Quality of Service (QoS) parameters:
 - CLR - cell loss ratio
 - CDV - cell delay variation
 - Max. CTD - Max. cell transfer delay
 - mean CTD - mean cell transfer delay

- user gives above parameter settings & route must be determined around them

Types of delay cell encounters:

 - PD - packetization delay at source
 - TD - transmission & propagation delay
 - QD - queuing delay at each switch
 - FD - fixed processing delay at each switch
 - DD - jitter compensation or depacketization delay at destination

- consider following example:
 - voice transmission - 64 kbps
 - transmission rate - 155Mbps
 - path length - 1000 km
 - path of 5 nodes

=> voice sampled at 8000 times/sec or once every 125 microsec.
- each sample in 1 byte => 125 x P micro sec, cell of P bytes
- so for cell of 48 bytes of data = 125 x 48 = 6000 micro sec.

e.g.   PD			6,000 microsec.
       TD			5,000
       PD			  280
       QD			   70
       DD			   70
 Total delay        	       11,420
  delay jitter         		   70


Note:  prop. delay approx. 5 microsec/km; trans. delay for cell is time for
 53x8 bits  at 155 Mbps or about 3 micro sec. so negligible compared to 5000

Statistical Multiplexing:


switch has 5 tasks: - demultiplexing,  routing through switch,  multiplexing, buffering, discarding

table  -> (VCIin,  input port,  VCIout,  output port)

 - routers in datagram NWs do not have such connection state information

- entry in routing table created at time of VC set up;  deleted when channel
 torn down

- switch uses statistical multiplexing for cell streams directed to same
 output port

Switching Techniques


circuit switching -> 2 techniques
 - space division switching - e.g. crossbar
 - time division switching - used for digital with signals put into time slots

- can use both techniques in same switch architecture 
 e.g. AT&T No. 4 ESS is  TSSSST  
       
	Fig. 12.5		Fig. 12.6            Table 12.1


3 types of architectural designs for fast packet switches recently:
 - shared-memory type
 - shared-medium type
 - space division type

Delta Network - subclass of Banyan that self-route (digit controlled)

			Fig. 12.21

rule for DN - if inputs to switching element come from other switching elements => both 
  inputs from upper lines of preceding
 -stage switching elements or both from lower lines


Photonic Switching


- expected that by early in next century, all telecommunications signals, long distance & 
   local loop, will be carried by optical fiber

- 3 main categories of switching systems that interconnect large collection of fiber-optic
  cables:

1) optical-to-electrical (o/e) conversion, elctronic switch (e.g. fast packet switch), electrical-
  to-optical (e/o) conversion

2) all optical (photonic) switches
 - avoid need (& cost, power) for o/e or e/o conversions
 - optical signal switched directly 
 - control of NW electronically implemented 
 - switch under electrical control; extremely wideband signals (e.g. Gbs) at potential 
   reconfiguration times on order < 1 ns

3) similar to 2)
 - no o/e or e/o conversion
 - control is photonic switch - optical

- current research on optical switches -> applications in space-, time-, & wavelength (freq) 
  division switching systems


Optical Time-Division Multiplexing

	Fig. 12.30

- optical source generates very narrow optical pulses (e.g. width of 1 to 10 ps, 
  corresponding to bandwidths of 100-1000 GHz

- narrow pulses split into N paths => for each path, narrow  pulses modulated by user's 
  data so narrow pulse will be passed (logical 1) or blocked (logical 0)

- delay inserted into path so that successive paths offset in time by one narrow pulse ->
  time multiplexing narrow bits associated with each user

=> composite split before reaching receiver & fed into optical AND gates -> optical AND 
  gates have 2 inputs:
  - composite signal
  - delayed replica of original narrow pulse stream

=> output of AND gates processed by optical receiver to electronically regenerate desired 
  packet


Wavelength-Division Multiplexing


     fig. 12.31

- each receiver assigned unique wavelength
- transmitter wishing to access given receiver tunes its transmitter to receiver's wavelength 
  & sends packets

- passive star coupler linearly combines all simultaneously transmitted packets =>
  assignment of different wavelengths to each packet preserves their individual identity,  but
  packets addressed to same receiver do collide

- drawback:  need bank of transmitting lasers, one for each receiver
 -> lasers should be rapidly tunable (able to tune in 10's of ns) over  broad optical band

- alternative -> assign unique wavelength to each transmitter
 => receiver selects from multitude of WDM signals, on packet-by-packet basis, 
  corresponding to wavelength at each point in time =>  side channel for signaling needed to 
  inform receiver of appropriate channel to which it must tune or rapidly tunable optical 
  filter used





Back to CS home page

CEN 5515 - Data Communications Notes

CEN 5515 - SONET Notes

CEN 5515 - ISDN Notes