CSNDSP06, Patras, Greece

1
On the Impact of Burst Assembly on
Self-Similarity at the Edge Router in Optical
Burst Switched Networks
ATHENS INFORMATION TECHNOLOGY

• Siamak Azodolmolky, Anna Tzanakaki, Ioannis
  Tomkos
• sazo,atza,itom_at_ait.edu.gr
• Athens Information Technology (AIT)
• P.O.Box 68,19.5km Markopoulo Ave., 19002 Peania,
  Athens, Greece
• Tel. [phone], Fax [phone]
• 5th International Symposium on Communication
  Systems,
• Networks Digital Signal Processing (CSNDSP06)
• July 19-21, 2006 Patras, Greece

2
Outline of Talk

• Introduction
• Optical Burst Switching (OBS) A Quick Overview
• Research Question
• Simulation Model
• Hybrid Burst Assembly
• Simulation Scenario
• Results
• Conclusion Future works
• Question Answers

3
OBS An Overview

• OBS is a transport scheme that combines the best
  feature of both circuit switching and packet
  switching
• Various types of client data are aggregated at
  the ingress node and transmitted as data bursts
• The Burst Control Packet is out-of-band and
  transmitted prior to the data burst
• At a lower rate than the data burst
• It is processed independent of the data burst in
  the electronic domain
• Time offset may be used for QoS differentiation
• The client data is buffered at the edge, where
  electronic buffer is cheap and abundant
• The data burst is transparently switched and
  routed through the network without OEO conversion

4
Burst Assembly (Aggregation)

• Timer-based (T)
• A timer starts at the beginning of each new
  assembly cycle. After a fixed time T, all the
  packets that arrived in this period are assembled
  into a burst.
• Burst size-based (L)
• Packets are assembled into a burst after a
  pre-defined length L is reached.
• Hybrid timer/threshold-based
• Its a technique, which combines both previous
  criteria

5
Research Question

• Traffic Shaping Capabilities of the burst
  assembly algorithms
• Goal Suppression of self-similarity level of the
  output traffic in the presence of self-similar
  input traffic
• Self-similar traffic sources
• Hybrid burst assembly algorithm
• Various Hurst parameter estimators are applied
• R/S plot, aggregate variance, Periodogram, local
  Whittle, Wavelet
• Focusing on packet-wise and burst-wise traffic
  measurements along with byte-wise measurements
• Important due to performance impact on control
  units in core routers
• Performance evaluation of the burst assembly
  algorithms
• Burst assembly delay
• Burst assembly delay jitter

6
OPNET Modeling Domains

• Modular and Hierarchical structure of OPNET

7
Developed Models

• Link models
• Based on OPNET point-to-point (PPP_adv) links
• Packet Format
• OBS Edge Router
• OBS Core Router
• Source Nodes
• Integration based on OPNET RPG traffic source

8
Source Nodes

• Standard raw packet generator (RPG) model
• Self-similar traffic model based on superposition
  of fractal renewal point process (Sup-FRP)
• Probability density function for interarrival
  times

9
Source Nodes

• Self-similar traffic traces (OPNET RPG model)
• Arrival rate 100 packet/sec., Fractal Onset Time
  Scale (FOTS) 0.1 sec., H0.6, 0.7, 0.8, 0.9

10
Burst Assembly Algorithm

• Hybrid Burst Assembly
• Timer Threshold
• Size Threshold
• ? and ? are parameters
• ELp is the mean IP packet length
• N Number of OBS egress routers
• M Number of sources

11
Simulation Scenario

• 12 Client networks are connected to the edge
  router via 10Gbps links (M12)
• Edge router is connected to the core via a 40Gbps
  link
• Traffic load (0.3, 0.5, 0.7)
• Client networks are divided in 4 groups(H0.7,
  0.75, 0.8, 0.9)
• N Destinations in core network
• N1, 10, 20
• Traffic monitoring 100 ?Sec.
• Simulation time 6 sec.
• Realistic IP packet size distribution (Average IP
  packet length 375.5 byte)

12
Results H vs. Traffic Load
13
Results H vs. ?
14
Results H vs. ?
15
Burst Assembly Delay vs. ?
16
Burst Assembly Delay Jitter
17
Burst Assembly Delay vs. ?
18
Burst Assembly Delay Jitter
19
Burst Creation vs. ? and ?
20
Conclusion

• Development of related OPNET models for
  performance evaluation of OBS networks
• A simulation framework using OPNET models
• Realistic self-similar traffic sources
• Performance evaluation of a typical network
  (assembly delay, delay jitter)
• Traffic shaping capability of hybrid burst
  assembly algorithm in terms of reduction in
  self-similarity level of output traffic
• Performance evaluation of hybrid burst assembly
  algorithm in terms of burst assembly delay and
  its jitter.