Optical Packet Format Issues and MAN Access Protocols

1
Optical Packet Format Issues and MAN Access
Protocols

• Gérard Hébuterne
• INT - UMR CNRS 5157 SAMOVAR
• gerard.hebuterne_at_int-evry.fr

2
Motivation

• Growth in bandwidth demand
• Optics already present
• Fibres, using WDM, offer Gbit/s

• Multiservices networks based upon processing of
  packets
• Why not all-optical packet switching?

3
However, all-optical packet switching raises
several issues

• Especially, no large-scale buffering available
• How to provide QoS, with small buffers, for
  links carrying Gbit/s or Tbit/s ?
• The spectral domain, a new paradigm to add to
  traffic engineering methods
• Specific behaviours
• Synchronization, buffers

4
Synchronization

• Absorb jitter encountered in fibres

Influence of temperature, etc.
Influence of Propagation
5
Optical buffers

• Electrical buffers
• the packet waits until available resource
• Optical buffers Fibre Delay Lines
• the packet waits for a fixed duration

duration FDL length
6
Optical buffers (2)
7
(A personal view of) The typical all-optical
architecture

• The MAN collects flows from LANs, to the core
  network

8
Packet format the choices (1)

• Fixed-length packets (FLP)
• packets processed/forwarded one at a time
• all packets have the same length

9
Packet format the choices (2)

• Variable-length packets (VLP)
• packets processed/forwarded one at a time,
• packets have arbitrary length

10
Packet format the choices (3)

• Slotted Variable-length packets (SVLP)
• packets processed/forwarded in bursts
• packets occupy an arbitrary number of slots

Header
Packet n
Packet m
Packet
11
Packet format traffic-related issues

• FLP, VLP, SVLP ?
• Which length (or minimum / maximum)
• Same format in WAN and MAN ?

• The decision must be based upon traffic
  arguments (and others!)
• efficiency in terms of bw consumption
• efficiency in terms of processing time

12
Current length of Internet packets
In nb of packets
In nb of bits
Source http//www.caida.org/analysis/AIX/plen_his
t/
13
A few figures

• Target bandwidths 10 to 40 Gbit/s
• Packet format
• duration several (up to 10) µs
• Guard bands headers synchronisation 50 to
  100 ns ?
• At 10 Gbit/s,
• 40 bytes (minimum) 30 50 ns
• 1500 bytes Ethernet frame 1.2 to 1.5 µs

14
Where is the choice significant ?

• At the network interfaces
• Client packets (ATM, Ethernet, etc.) have to fit
  into optical containers.
• At the MAC level
• Performance of simple/fast mechanisms
• In the core network
• Priorities, deflection, etc.
• Hardware aspects
• Standardisation

15
The Network interface

• LO size of fixed length optical pckt
• MO max size variable-length optical pckt
• ME max size electrical pckt (e.g. Ethernet)

Fixed
Variable
MO gt ME
LO gt ME
LO lt ME
MO lt ME
Aggregation mandatory
Segmentation mandatory
Segm/Reass mandatory
Nothing to do!
Additional delays ? Bandwidth loss ?
16
Example aggregation

• Need of a time-out to upper limit the aggregation
  delay

Packet arrival
Timeout! Packet sent
optical packet
17
Efficiency of aggregation

• Measured in delay and fill ratio

Packets wait until completion of the optical
packet or time-out fires
Optic
Electronic
10 Gbit/s
18
Aggregation

• Arriving packets are stored in the optical
  container
• As soon container full (i.e. the arriving packet
  too large), send the container
• If it takes too long, send the container
• (the time out begins at the container is created)

19
Analysis (no time-out)

• Xn the size of the container after n-th packet
• fn size of n-th packet
• K size of container

20
Efficiency of aggregation
Client flow 1 Gbit/s
Client flow 0.5 Gbit/s
21
Aggregation and delays

• Obviously, performance level (delay, efficiency)
  depends on
• the bitrate of the aggregated flow
• the length of the optical packet
• the time-out value

22
In the Core Network

• How to perform prioritization mechanisms related
  with QoS provision ?
• What about contention resolution in switching
  stages ?
• Importance of synchronization

23
Contention resolution

• Deflection routing, eulerian circuits ?
• Use of buffers (of limited size)
• Case of synchronized, FLPs QoS levels handled by
  priorities
• Non-synchronized or VLPs priority schemes
  difficult to handle

24
MAN Access protocol

• Assume the simplest access protocol empty
  slot protocol
• Hopefully, if load level not too high, fairness
  will not be an issue ?
• If needed, however, a more sophisticated control
  can be implemented
• Then, how to compare FLP and VLP ?

25
With variable length packets

• Gaps may form on the ring, as the packets
  progress
• lowering the capacity effectively available
• increasing unfairness

26
A model for variable-length packet insertion

• The transmitter observes the fibre in order to
  detect a gap longer than the packet to send

27
Model (contd)

• An equivalent model
• instead of waiting for a gap large enough, the
  transmitter sends its packet, but interrupts it
  as soon as a packet arrives on the fibre
• ? Equivalent to the Preemptive repeat-identical
  priority model !

28
To illustrate
29
Example

• 20 stations,
• ring _at_ 10 Gbit/s
• 500 Mbit/s per station

30
Concluding remarks

• Specificity of all-optical packet networking
• Also, uncertainty in the assumptions of any
  traffic study
• services actually offered
• distribution of electrical packet lengths
• real demand on different QoS levels

31
Concluding remarks (2)

• Packet format remains an important open issue,
  esp. because a standardized format conditions
  future developments
• Different formats in access and core ?
• Other (traffic related) arguments to be developed
  ?
• Etc.

32
Concluding remarks (3)

• Other ways are considered
• optical burst switching (kind of fast circuit)
• circuit switching in the core !

See, e.g. http//www.stanford.edu/nickm/talks/Op
enArch_2003.ppt
33
Thank you

• for your attention