TomTat(Tieng anh)

When the control pack of a burst arrives at a network core node, based on the information contained in the control packet as the arrival time, the end time of the burst, a scheduling alg[r]

NGUYEN HONG QUOC

RESEARCH ON SOME SCHEDULING METHOD

IN OPTICAL BURST SWITCHING NETWORK

MAJOR: COMPUTER SCIENCE

CODE: 62.48.01.01

SUMMARY OF THESIS COMPUTER SCIENCE

Supervisors:

1 Assoc Prof Dr Vo Viet Minh Nhat

2 Dr Nguyen Hoang Son

HUE, YEAR 2017

1 Assoc Prof Dr Vo Viet Minh Nhat Hue University, Viet Nam

2 Dr Nguyen Hoang Son Hue University of Sciences, Viet Nam

The dissertation will be defended at University level Council of Dissertation Assessment at Hue University

Time: date month year 2017

Fiber-optic networking since its inception in the 1990s has gone through many ations of development: from early wavelength-routed (WR) models based on early light-paths end-of-life, up to the proposed Optical Packet Switching (OPS ) models, with theidea coming from the packet switched network model However, with some technologylimitations, such as the production of optical buffers (something like analog RAM in theelectronic environment) or micro-speed switches, the packet switching model can not yet

gener-be gener-become reality A proposed compromise solution called Optical Burst Switching (OBS )has opened the way for new research and is considered a promising technology for thenext generation of the Internet

A typical feature of an optical burst swtiching (OBS network ) is the Burst HeaderPacket (BHP ), which is separated from the Data Burst (DB) segment In other words, toimplement a burst transfer into the core network, an BHP control packet is generated andsent before an offset time This offset should be calculated sufficiently to place resourcesand configure switches at intermediate nodes along the path of the optical burst from thesource to destination However, this method of transmission also puts pressure on how

a BHP control pack schedules resource scheduling and switch configurations at the corenodes, ensuring subsequent burst transmission That is the task of scheduling resourcereservation at network core nodes Therefore, scheduling issues need to be addressedand researched to maximize bandwidth performance, reduce data loss and improve theperformance of OBS networks

2 Research Motivation

Scheduling is one of the most important activities in optical burst switching networks.When the control pack of a burst arrives at a network core node, based on the informationcontained in the control packet as the arrival time, the end time of the burst, a schedulingalgorithm will now be called to find the step channel Outgoing waves are available toschedule incoming bursts The main purpose of the scheduling algorithm is to arrange theincoming bursts on wavelength channels to maximize the bandwidth utilization, reducethe number of bursts removed, and improve the performance of the OBS network Thereare already many proposed scheduling algorithms that can be categorized into two mainapproaches:

• Online method of scheduling

• Group method of scheduling

For online scheduling when a control packet reaches a network core node, one of theonline scheduling algorithms will be called immediately to find the available wavelengthchannel for its burst; If more than one wavelength channel is available, this schedulingalgorithm will select a scheduling channel that optimizes the algorithm’s set point Amongthe online scheduling algorithms, BF-VF is the best scheduling algorithm for bandwidthutilization However, the efficiency of online scheduling depends on the bandwidth status

that combine online scheduling with re-scheduling and burst segmentation have beenproposed Specifically, when the online scheduler does not find the available channel, in-stead of the incoming burst it will be dropped completely, rescheduling will rearrange theburts that has been scheduled on wavelength channels to search for the bandwidth loca-tion Seek to schedule incoming bursts or perform segmentation to drop only a portion

of the burst segment bursts However, online scheduling and online scheduling combineonly optimal bursts scheduling to the present without regard to the subsequent burst.Bandwidth decomposition is created by current burst scheduling and can affect the effi-ciency of subsequent scheduling The group scheduling method is therefore proposed inwhich the control packets arrive over a period of time that will carry out the scheduling

of their corresponding burst Depending on whether the network core node is equippedwith a full wavelength converter, group scheduling algorithms can be divided into twogroups: Group scheduling on a single channel in case of non-use of wavelength convertersand group scheduling on multichannel when fully equipped with wavelength converters.However, the above scheduling algorithms still reveal the following shortcomings:

• Combined scheduling algorithm: Do not use the best online scheduling rithm in phase 1 to schedule incoming bursts The re-scheduling of phase 2 algo-rithms is only considered for the last burst on the output channels The overlapusing the phase-3 burst segmentation was eliminated

algo-• Group scheduling algorithm single-channel: Computational complexity of highalgorithms; Not taking advantage of the bandwidth void filling created between thescheduled burst for scheduling incoming bursts and scheduling slots are set at afixed value that does not care for incoming bursts traffic

• Group scheduling algorithms on multi-channel: Heuristic-based algorithmshave not yet set the optimum scheduling option for incoming bursts, but only based

on sort order The optimal scheduling algorithm increases the number of controlpackets, requiring system changes to be made Furthermore, removing all the burstsscheduled on the channels to bring about the scheduling problem on the host ma-chine is unrealistic on the real network

The above-mentioned shortcomings are the motivation for the thesis to focus on searching, improving and proposing new scheduling algorithms to minimize loss of data,maximum bandwidth usage, Reduce computational complexity and improve network per-formance

re-3 Research objectives of the thesis

The goal of the thesis is to study, improve and propose some scheduling algorithms

to improve the performance of an optical burst switching network including: minimumdata loss, maximum bandwidth efficiency, decrease the delay and reduce computationalcomplexity The specific objectives of the dissertation are:

• Research, improve online scheduling algorithms in combination with re-schedulingand burst segmentation

• Research, improve and propose new group scheduling algorithms on single-channel

• Research, improve and propose new group scheduling algorithms on multi-channe

Based on the research objectives, the dissertation is implemented in three main researchareas:

• Problem 1 : Improve algorithm that combines online scheduling with re-schedulingand burst segmentation

• Problem 2 : Improve and proposed new group scheduling algorithm on single-channe

• Problem 3 : Improve and proposed new group scheduling algorithm on multi-channel

4 Contribution of thesis

The main contributions of the thesis include:

• Propose algorithm for online scheduling, re-scheduling and segmentation of iCSA[CT2]

• Propose group scheduling algorithm on LGS single channel [CT8] and LGSVF [CT4],LAGS [CT5], LAGS-VF [CT7]

• Propose algorithm for group scheduling on multichannel OPT-GS in the optimalapproach and LGS-MC [CT6], LGS-MC-VF [CT3], MWC-GS [CT1], MWC-VF-GS[CT1] towards a heuristic approach

CHAPTER 1: OVERVIEW OF SCHEDULING IN

OPTICAL BURST SWITCHING NETWORKS

1.1 Brief history of optical communications

1.2 Optical switching models

Optical switching models can be divided into three types: optical circuit switching,optical packet switching and optical fiber switching

1.2.1 Optical circuit switching

1.2.2 Optical packet switching

1.2.3 Optical burst switching

1.3 Optical burst switching

OBS is considered as a promising technology for the next generation of optical internetbecause it has more functions and advantages than other optical switching networks.Some features of the OBS network are: separation between the control packet transmissionchannel and the burst channel; Dedicated one-way; burts length varies; No optical buffer

1.3.1 OBS Network Architecture

An OBS network consists of optical switch nodes (OBS nodes) connected by opticalfibers Each fiber is capable of supporting multi-wavelength channels There are two types

of OBS nodes: Edge node and core node

1.3.1.1 Edge node

1.3.1.2 Core node

1.3.2 Activities within the OBS Network

within the OBS network include: burst assemble, signaling, scheduling, and contentionresolution Each activity plays an important role and directly affects the performance ofthe OBS network

1.3.2.1 Assemble

1.3.2.2 Signaling

1.3.2.3 Scheduling

1.3.2.4 Routing

1.4 Scheduling in OBS networks

1.4.1 Introducing the problem of scheduling

1.4.2 Some related knowledge

1.4.3 Scheduling algorithms have been announced

Scheduling is one of the most important activities in OBS When a control packet rives at a node, depending on the destination of the corresponding burst, the resource will

ar-be reserved at the output port, including the wavelength channel and the hold time to ar-beallocated There are several recommended scheduling algorithms for different approaches

to improve the efficiency of scheduling These scheduling algorithms can be categorized

as online scheduling, online scheduling combined with re-scheduling and segmentationtechniques, and group scheduling

1.4.3.1 Online scheduling

When a control packet arrives at a node, a scheduling algorithm is called to schedulethe corresponding burst on a data channel Based on the information in the packet, thescheduler knows the arrival time, burst length, and conducts channels finder to schedulethe burst

online-scheduling algorithms in OBS can be divided into two categories: schedulingdoes not void filling and scheduling void filling

1.4.3.2 Online scheduling combination

Online scheduling algorithms can fail when no channel is available to schedule incomingburst As a result, the bursts will be eliminated and will cause large data loss, while burstscheduled on outgoing channels can be rearranged to create free bandwidth to be able

to schedule incoming burst In the case of inevitable elimination, burst segmentationtechniques can reduce the loss if the overlap is eliminated and the rest of the burst isscheduled An integrated approach is therefore proposed to avoid or reduce the removal

of whole burst The following are the combined approaches that have been announced

1.4.3.2.1 Combining online Scheduling and rescheduling

Integration The idea of scheduling is to rearrange resources for burst scheduled onwavelength channels so that the available bandwidth is available Play for the followingburst The purpose of scheduling is to increase bandwidth utilization on outgoing channels,reduce burst losses, and limit other complex processing

1.4.3.2.2 Combining online Scheduling and burst segmentation

The burst segmentation concept was first proposed by Vokkarane et al., The burstbeing divided into segments so that in the event of a conflict, only a few segments wereremoved and segments The rest is still transmitted through the network The missingsegments can be head dropping or tail dropping

1.4.3.2.3 Combining online scheduling, rescheduling and burst mentation

seg-Not only combines online scheduling with segmentation, Umaru, Aydin, and others pose another combination of scheduling, scheduling, and segmentation Specifically, thePCSA (Preemptive Channel Scheduling Algorithm) is based on LAUC-VF, ODBR andburst segments This combined scheduling scheme is designed to meet QoS requirements

pro-in the OBS network Another combpro-ination, SODBRA (Segmentation-based On-DemandBurst Rescheduling Algorithm), is a combination of FFUC-VF, ODBR, and burst seg-ments

wave-1.4.3.4 Combining scheduling and contention resolution

1.4.3.4.1 FDL

1.4.3.4.2 Wavelength conversion

1.4.3.4.3 Deflection routing

1.4.4 Some annotated scheduling algorithms

The scheduling algorithms have been published in different approaches to reduce theprobability of data loss, optimizing bandwidth performance However, there are still someshortcomings of this model, the scheduling algorithm So further improvements, new pro-posed models and scheduling algorithms are needed to better utilize bandwidth, reducethe probability of packet loss, and improve the efficiency of OBS scheduling

1.5 Summary chapter 1

In this chapter, the thesis introduces an overview of the optical switching network aswell as the activities within the network, highlighting the advantages of OBS over otherswitching networks The thesis then presents the scheduling algorithms in terms of onlineapproach, online scheduling, combined online scheduling, single-channel scheduling, andmulti-channel group scheduling Finally, the thesis gives the advantages and disadvantagesand hence the basis for further improvement and proposed new scheduling algorithms toimprove the performance of the network, reducing the probability of loss data, maximumbandwidth usage, reduced latency, and reduced computational complexity

CHAPTER 2: A COMBINED IMPROVEMENT

FOR ONLINE SCHEDULING WITH RESCHEDULING

AND BURST SEGMENTATION

2.1 Introduction

Online scheduling algorithms may fail to schedule a burst if the bandwidth is notavailable at the output channels However, scheduling can be accomplished if there is areordering for the bursts scheduled on the output channels, so that the bandwidth createdcan schedule the incoming burst This is the idea of ODBR and ABR But in the case ofburst removal, the burst segment is a suitable solution to reduce data loss, in which onlythe overlapping burst is removed while the other burst is formed Schedule to forward

to the next node A complete integration, including online scheduling, re-scheduling, andburst segmentation, such as PCSA and SODBRA, has been proposed to make schedulingmore effective for incoming burst

2.2 Analysis and evaluation of published scheduling algorithms

The combined scheduling algorithms can be classified as follows:

• Combination online scheduling with rescheduling including ODBR and ABR ing

schedul-• Combination online scheduling with burst segmention including MOC, MOC-VF

NP-• Combination online scheduling, rescheduling and burst segmentation including PCSAand SODBRA

2.3 The iCSA proposed scheduling algorithm

iCSA (Improved Composite Scheduling Algorithm) was announced in [CT2] iCSA is

a combination online scheduling, re-scheduling and burst segmentation

It consists of three phases that are implemented sequentially depending on the success

or failure of the previous phase

Stage 1: When a burst arrives, the BFVF void fill scheduling algorithm is called tofind the best fit available for scheduling this burst If successful iCSA scheduling ends,reverse phase 2 is called

Stage 2: Reschedule based on an improvement of the ODBR algorithm For burstthat were previously scheduled on outbound channels with overlapping burst so thatthe generated gaps could be scheduled for incoming burst In cases where more than onechannel is satisfied, the channel with the least number of bursts will be selected to schedulethe incoming burst If the schedule is not successful, Phase 3 will be called

Phase 3: Calculate the burst overlap to the burst scheduled on the channels andchannels with the smallest overlap selected to schedule the incoming burst after the overlaphas been cut; If the overlapping part can be scheduled on one of the remaining channels,

a new burst will be created and scheduled on the available channel (assuming the OBScore node is capable of creating the control packet)

The iCSA scheduling algorithm represents as follows:

Algorithm 2.1: iCSA (improved Composite Scheduling Algorithm)

Input : ub(sub, eub), W = {1, 2, , m}, SBk với k ∈ W ;

7 if (sub ∈ [si,k, ei,k]) ∨ (si,k ∈ [sub, eub]) then

8 OBk := OBk∪ {sbi,k}; refi,k := 0;

22 Rescheduling burst sbi,k on channel k to channel refi,k;

23 Resend control packet for burst sbi,k;

24 Remove the burst sbi,k schedule on channel k ;

25 Schedule burst ub on channel bestchannel and return;

44 eoverlap:= eub; soverlap:= sub− bestlenght;

45 Create a new burst from the overlapping segment newub(soverlap, eoverlap);

46 ch := BF V F (newub, W );

47 if (ch 6= −1) then

48 Schedule burst newub on channel ch;

49 Create a new control packet for the burst newub và return;

50 if (droptype = 1) then

51 sub := sub− bestlenght;

52 Schedule burst ub after removing the overlapping segment on the channel

bestchannel;

53 eoverlap:= eub; soverlap:= sub− bestlenght;

54 Create a new burst from the overlapping segment newub(soverlap, eoverlap);

55 ch := BF V F (newub, W );

56 if (ch 6= −1) then

57 Schedule burst newub on channel ch;

58 Create a new control packet for the burst newub và return;

2.4 Simulation and analysis of results

To demonstrate the effectiveness of the proposed algorithm, the proposed tation of the proposed iCSA scheduling algorithm simulates the published algorithms,including LAUC, BFVF, ODBR, ABR, SODBRA and PCSA The simulation environ-ment uses the NS2-obs0.9a package and the C ++ software, installed on the 2.4 GHzIntel Core 2 CPU, 2G RAM Network model simulates NSFNET with 14 core nodes, witheach edge node connected to a core node

implemen-Based on the simulation results shown in Figures 2.8 to 2.15, when comparing theproposed iCSA scheduling algorithm and published scheduling algorithms, the probability

of packet loss of the iCSA algorithm is much lower The number of control packets to besent back due to segmented burst and algorithmic complexity equal to those of the sametype

Figure 2.8 Compare based on

packet loss probability between

LAUC and BF-VF

Figure 2.9 Compare based on packet loss probability between ODBR and ABR

Figure 2.10 Compare the number

of bursts must be rescheduled between ODBR and ABR

Figure 2.11 Compare based on packet loss

probability of iCSA versus ODBR and ABR

Figure 2.12 Comparing bursts must be re-scheduled of iCSA versus ODBR and ABR

Figure 2.13 Compare based on packet loss probability of iCSA

versus SODBRA and PSCA

Figure 2.14 Comparing bursts must be re-scheduled of

iCSA versus SODBRA and PCSA

Figure 2.15 Comparing bursts must be segmented bursts

of iCSA versus SODBRA and PCSA

2.5 Summary of Chapter 2

This chapter of the thesis presents, analyzes, and evaluates in detail the scheduling rithms that combine online scheduling and ODBR scheduling, ABR and online schedulingcombined rescheduling and segmentation techniques Burst were announced SODBRA,PSCA Based on analyzing, evaluating and presenting the advantages and the existence

algo-of algorithms, the thesis proposes a online scheduling algorithm combining re schedulingand burst segmentation techniques With the advantages:

• iCSA proposed algorithm using online scheduling algorithm BFVF is the best ing algorithm in online scheduling algorithms for scheduling incoming burst

schedul-• In Phase 2, iCSA performs a rescheduling of any overlapping burst with incomingburst, thus reducing the probability of packet loss and better utilization of band-width

• In Phase 3, the iCSA finds space (available bandwidth) on a channel to schedulethe overlap This will reduce the probability of packet loss and better bandwidthutilization

The simulation results demonstrated that the proposed iCSA scheduling algorithm has

a lower packet loss probability than previously published algorithms and has algorithmiccomplexity equal to that of the same algorithm

CHAPTER 3: SOME IMPROVEMENTS OF GROUP SCHEDULING ALGORITHMS ON SINGLE CHANNEL

3.1 Introduction

Group scheduling at the OBS core node is the operation that the control packetsarrive in each time slot τ , based on the information in the control packets A schedulingalgorithm is called to conduct simultaneous scheduling for their respective bursts (asshown in Figure 3.1) Group scheduling can be performed on only one output channelwithout the support of wavelength conversion and these incoming bursts are assumed tohave the same wavelength This type of scheduling is called group scheduling on a singlechannel Where group scheduling is supported by wavelength conversion, called groupscheduling on multi-channel will be presented in the next chapter

The next chapter of this chapter summarizes the previously published single-channelscheduling algorithms Based on analysis and evaluation of strengths and weaknesses, thethesis proposes a number of multi-channel scheduling algorithms to improve schedulingefficiency Simulation and analysis setup results are also described in detail in this chapter

Figures 3.1 (a)Describe the operation of group scheduling on a single-channel and

Figure 3.6 The proposed activity model for group scheduling on a single-channel

Phase 1: Classification of arrival packets in the time slot τ based on the incomingwavelength of the burst The information in the packet includes the arrival time, the endtime and the incoming wavelength channel of the corresponding burst Each wavelengthchannel retains a LAUT (Lastest Available Unscheduled Time) value which is the lastavailable time that can be scheduled for incoming bursts A packet arriving in the timeslot τ is considered eligible for scheduling if the arrival time of the corresponding burst isgreater than the LAUT value on its output channel Based on the incoming wavelength