HUE UNIVERSITY UNIVERSITY OF SCIENCES PHAM TRUNG DUC
RESEARCH SOLUTIONS TO IMPROVE QUALITY OF SERVICE IN OPTICAL BURST SWITCHING NETWORK
MAJOR: COMPUTER SCIENCE CODE: 9.48.01.01
SUMMARY OF PHD THESIS
HUE, YEAR 2021
The thesis has been completed at: ................................................................. ........................................................................................................................ ........................................................................................................................ Supervisor: 1. Assoc. Prof. Dr. Vo Viet Minh Nhat, Hue University 2. Dr. Dang Thanh Chuong, University of Sciences, Hue University ........................................................................................................................ Reviewer 1: ..................................................................................................... ........................................................................................................................ ........................................................................................................................ ........................................................................................................................ Reviewer 2: .................................................................................................... ........................................................................................................................ ........................................................................................................................ ........................................................................................................................ Reviewer 3: .................................................................................................... ........................................................................................................................ ........................................................................................................................ ........................................................................................................................ The thesis will be presented at the thesis Committee meeting of Hue University, to be held by Hue University at: ................................................. …………………….................. ...................................................................... The thesis can be found at the following libraries: ........................................ ........................................................................................................................ ........................................................................................................................ ........................................................................................................................
PREFACE
1. The urgency of the topic
With the explosion of network applications in recent years, the transmission of data over the network has become a challenging problem and it is attracting a lot of attention. There have been many different proposals about data transmission methods, from traditional information transmission way via copper cables, over radio waves to burst swich support multichannel transmission, in which optical bursts have many advantages such as low attenuation, very large bandwidth and immunity to electrical interference compared to copper cables. With the recent great success of Wavelength Division Multiplexing (WDM) technology, the bandwidth of each optical burst is separated into multiple wavelength channels, thereby responds better increasing communication needs of the users [35], [53].
Optical communication, since its inception until now, has gone through three generations of development, from the original WR wavelength routing models providing point-to-point links, to the second generation with end-to- end optical links (lightpath) which is reserved in the optical layer. In the 3rd generation, Opitical Packet Switching (OPS) models [51] are proposed with the idea which is inspired by the packet-switched network. In order to deploy on ring or grid topologies to be able to adjust flexibly in responsing to change of the flow rate. However, with some technological limitations, such as not being able to produce optical buffers (similar to RAM in electrical networks) or optical packet switches at nanoseconds rate, Optical Packet Switching (OPS) can’t be come reality. One compromise solution is the Optical Burst Switching (OBS).
A typical feature of communication in the Optical Burst Switching (OBS) network is that the BCP control part (packet) is separated from the DB data part (burst). In other words, in order to transmit an optical burst, a control packet is formed and is sent before an offset (offset time). This offset time should be calculated such that it is sufficient to preset the resource and configure the switches at the intermediate nodes throughout the journey which the optical burst will pass from the source node to the destination node. Not only is it separated in terms of time, the BCP control packet is also separated in terms of space from its data burst, where several channels (wavelengths) are reserved for the BCP control package, while the other channels is used for data burst transmission [75].
With this data transmission, it is clear that the OBS network does not need optical buffers to save temporarily data bursts while waiting for the processing their BCP control packets at the intermediate nodes (the core
1
node) and the OBS network also does not require nanosecond rate switches. However, this communication way also puts a pressure. That’s how a control package can reserve resources promptly and switching configure success at the core nodes, which ensure optical burst transition follow it. That is the task of such activities as reserving resources, scheduling and disputing resolution [19].
A solution to enhance service quality in the OBS network can be performed by providing differentiation of service quality at some point (node) in the OBS network [32]. Specifically, typical approaches to mechanisms that provide this distinction can be: differentiation at the control layer and data layer [45], where the activities provide differentiation of time, service quality can be: differentiation about compensation differentiation in dispute resolution policy, differentiation in the process of gathering burst and differentiation in some scheduling activities ... [32]. These models are essential to have effective control mechanisms to provide the QoS differentiation which is committed, and can provide more resources for different applications to optimize communication performance on the whole network (based on requirements about latency, data loss rate and bandwidth constraints ...). 2. Research motivation
There are available studies to improve quality of services (QoS) in the OBS network, which can be classified into 2 main approaches solution groups:
- Improve QoS at the edge node; - Improve QoS at the core node. The above solution groups usually aim to improve QoS through scheduling admission control process [5], [33] at the core node, provide differentiation of QoS at the edge node, or improve QoS on the nodes. 3. Research objectives
• Research and improve scheduling admission control mechanism to enhance QoS based on incoming burst rate prediction at core node to improve scheduling efficiency for low QoS bursts but still ensure a level of service quality for high QoS bursts. The effectiveness of the scheduling admission control mechanism was valued through simulation and mathematical analysis.
• Research and propose scheduling historical data anylysis method to identify factors affecting to scheduling efficiency, thereby proposing solutions which reduce data loss to improve scheduling performance at core node. • Research and improve the mechanism which provide difference of based
2
on offset time and burst size at the egde node. Based on the available resource information base which is reflected from the core node, the egde node to adjusts the size of the burst is generated to provide bandwidth efficiency, reduce burst loss rates but still improve QoS for each priority class.
4. Object and scope of the study - Object of study: Models and admission control algorithms and burst aggregation in the OBS network.
- Scope of study: Edge node and core node in OBS network. 5. Research Methodology - Theoretical research method: Synthesize of publications related to models, algorithms, improved diversity assurance mechanism and provide QoS. Analyze and evaluate the pros and cons of published proposals as a basis for improvement or new proposal. - Simulation and experiment methods: Install innovative algorithms and propose new solutions to prove the correctness of these algorithms.
6. Thesis structure The thesis includes the introduction, three chapters of content, the conclusion and the list of references. Specifically:
- Chapter 1, with the chapter title "Overview about quality of service in optical burst switching networks", presents the basic knowledge about optical burst swithching networks, including: history of optical communications, optical switching models, optical-burst switching network’s architecture, network internal operations and the problem of improving QoS on the OBS network.
- Chapter 2, with the chapter title "Solutions to improve service quality at the core node", focuses on the main issue: proposing some models which are used to predict the incoming burst rate based on admission control.
- Chapter 3, with the chapter title "Solutions to improve service quality at the egde node and the combination of nodes", solves (3) the problems including: (1) presentating synthesis of the research related to the mechanism which provide QoS at the edge node, (2) finding out the cause of the burst loss based on analyzing scheduling historical data to improve performance at the core node and (3) reviewing the control packet structure to propose the QoS providing model on the whole network after receiving feedback on the gap at the core node which is sent back in the control packet to adjust the burst aggregation phase in order to optimize the used bandwidth and provide efficiency in the problem how to improve QoS.
"Conclusion and development direction of the thesis" stated the contributions of the thesis, development direction and concerns of the author.
3
CHAPTER 1. OVERVIEW ABOUT QUALITY OF SERVICE IN OPTICAL BURST SWITCHING NETWORK 1.1 Introduction about optical burst switching network Optical switching is divided into 3 types: optical channel switching,
Quality of service in the OBS network
resource
optical packet switching and optical burst switching. 1.1.1 Architecture of the OBS network 1.1.2 Comparison of optical switch models 1.1.3 Operations at the edge node 1.1.4 Operations at the core node 1.1.5 Scheduling in OBS network 1.2 1.2.1 The need to improve quality of service (QoS) in the OBS network QoS enhancement can be done by some mechanisms such as providing/improving QoS at each node or combining nodes in the OBS network, in order to create a variety of options which provide services at to achieve the required service, original request. According to [39], depending on the model, mechanism or technique, the enhancement of the QoS provisioning/improvement mechanism can be classified into two main ways, they are the QoS improvement mechanism at the nodes and the QoS provisioning mechanism. The solution enhances the QoS improvement mechanism which is implemented on all nodes, based on mechanisms to improve the general performance of the network, specifically: burst aggregation, signaling scheme reservation, scheduling for algorithms and disputing resolution. Then, the QoS providing mechanisms will be considered, which means that they concern to the QoS label (high priority burst or low priority burst) with different approaches [76]. Another QoS enhancement solution can achieve by the typical approach in differentiation mechanisms which are implemented at the control or data plane [45], where operations to provide QoS differentiation can be: signaling and routing at the control layer. At the data layer at the edge node, there are mechanisms that provide QoS differentiation through parameters during aggregation process, burst size ..., the core node has control models to enhance the QoS differentiation from the egde node related to dropping or scheduling policies. 1.2.2 1.2.3 Improve quality of service at the core node Improve quality of service at the egde node
4
1.3 The research objectives of the thesis
Based on the analysis of published solutions about enhancing QoS in OBS network and the general research goal identified in the introduction section, specifically proposing some solutions to improve QoS in OBS network. The thesis is implemented under 4 main research objectives, including:
• Objective 1. Propose scheduling admission control method based on incoming burst rate prediction in order to improve QoS provision at the core node. • Objective 2. Research and improve the quality of service differentiation mechanism at the egde node based on the combination between burst length adjustment and offset time. • Objective 3. Research and propose methods of analyzing scheduling historical data in order to identify factors affecting to scheduling efficiency, thereby proposing solutions which reduce data loss and improve scheduling efficiency. • Objective 4. Study and propose the QoS enhancement model by associating at the edge node and the core node. In which, the core node is responsible for responding to the gap information, the edge node adjusts the burst aggregation process to increase successful scheduling rate but still improve QoS differentiation. In which, objective 1 will be implemented in Chapter 2 and objectives 2,
3, 4 will be presented in Chapter 3. 1.4 Summary of Chapter 1
The first chapter of the thesis introduced an overview about the OBS network and operations inside the network, in which the scheduling admission control problem at the core node and the offset time differentiation at the edge node which is focused on analysis because it has an important effect to the QoS enhancement problem on the whole network. The thesis also analyzed and evaluated the published methods so far about admission control and differentiation QoS. That is the basis for the final thesis to identify 4 goals which need to be studied as well as propose functional modules which are added to enhance QoS deployment and communication performance of OBS network.
5
CHAPTER 2. SOLUTIONS TO IMPROVE QUALITY OF SERVICE AT THE CORE NODE
2.1 Scheduling admission control is supported to provide quality of service
Scheduling admission control (admission control) can be deployed at each output port of egde node and core node. Thanks to the use of electronic buffers, scheduling admission control at the egde node is often simpler and more efficient. 2.2 Analysis and evaluation of admission control models 2.2.1 Wavelength group model 2.2.2 Comparison and evaluation based on simulation
The thesis conducted the effective comparison between the scheduling admission control models of SWG, DWG and LLAC based on simulation which is installed with the supportion of NS2 network simulation software [77] integrated with Obs 0.9 package, on a PC with CPU 2.4GHz Intel Core 2.2G RAM. The simulation network is NSFNET. a. Comparing burst loss rate
DWG's total burst loss rate was the lowest (Figure 2.4c). LLAC always gives priority to high priority bursts so it always get the lowest high priority burst loss rate, but LLAC with low priority burst loss rate is quite high. This is true with setting result which is performed in [22].
Figure 2.4 Comparison of burst loss rates of high and low QoS class between SWG,
DWG and LLAC
b. Comparing bandwidth utilization
When considering bandwidth utilization rate, DWG has the highest bandwidth utilization, for both 2 classes (Figure 2.5). This is understandable when DWG reduces the number of loss burst it will increases bandwidth utilization efficiency.
6
Figure 2.5 Bandwidth utilization in 2 classes of the SWG, DWG and LLAC model
2.2.3 Result 2.3 The admission control model is based on the ARP-SAC incoming burst rate prediction 2.3.1 The prediction model is based on the incoming burst rate
Let is the arrival rate of the high priority bursts and is the arrival rate of the low priority bursts, the number of channels assigned for the low priority can be calculated in proportion to the arrival rates of the two types of high priority and low priority burst which are shown in formula (2.2). Note that high priority bursts are used all channels, it means …
is the average arrival rate in the past,
(2.2) ⌈ ⌉ Predictive model based on TW-EWMA method [32] determines the incoming burst rate by equation (2.6):
and
which
(2.7) The scheduling admission control model based on adaptive arrival rate
(2.6) is the current . In [3], this arrival rate, and is the weight of weight is chosen as . However, as the proposal in [60], this coefficient can be adjusted flexibly based on average rate in the pass and current rate as equation (2.7).
prediction therefore it’s called ARP-SAC model. 2.3.2 Describing an admission control algorithm in the ARP-SAC model
Algorithm 2.1:
Input:
-
Set
bursts
each
coming ,
inside and time is coming and ending,
7
defines as high (0) or low (1) QoS.
//output wavelength channels
- ;
Output:
- ; - Set of high QoS burst is scheduling to , and dropped ; - Set of high QoS burst is scheduling to , and dropped .
Process: 1
;
2
3
;
4
; ; ;
5
;
6
while ( ) do
7
chùm đầu tiên từ tập ;
8
;
9
if ( – ) then
10
if ( ) then
11
;
12
else
13
;
14
end if
15
else
16
if ( ) then
17
;
;
18
;
;
19
;
20
;
⌉;
21
⌈
22
; ;
23
;
24
end if
8
25
end if
26
if – then
;
27
28
;
end if
29
30
if (( ) ( )) then
;
31
32
if ( ) then
33
;
34
else
35
36
end if
37
else
38
if ( ) ( ) then
;
39
40
end if
41
if ( ) then
42
;
43
else
44
;
45
end if
46
end if
47 end while
2.3.3 Comparing and evaluating based on simulation results The thesis uses the parameters which are set in this part as Section 2.2.2. a. Comparing burst loss rates
The result in Figure 2.9c shows that the burst loss rate of ARP-SAC always gives the best results thanks to the policy of allocating more bandwidth for the low priority burst traffic. The improvement about the burst loss rate for the low priority burst in this case is 44%, 38% and 20% in the cases of incoming traffic rate is 3:7, 5:5 and 7:3 respectively.
9
Figure 2.9a Comparison of low QoS class burst loss rate between SWG, DWG, LLAC and ARP-SAC b. Comparing the variation about the number of wavelengths allocated for the low priority class.
Figure 2.10 Comparison of wavelength allocation for low priority burst traffic
2.1.3.4 Result
Based on simulation results, the admission control model based on ARP- SAC icoming burst rate predictions shows low priority burst loss rate and total loss rate decrease about 30% and 15% compared to others models . However, in terms of the high priority burst loss rate, the ARP-SAC method has burst loss rate which is 3% higher compared to the previously published models (Figure 2.9b).
This is because the ARP-SAC model reserve the number of allocated wavelengths for the flexibly low priority burst follows the incoming burst rate without using a minimum wavelength value, so the result is that in a low priority burst loss rate reduction, and lead to the number of wavelengths that provide for the high priority class are occupied; this increases priority of
10
burst loss rate, though not much. To solve this problem, the thesis continues to propose a new rate prediction model which will be presented in the next section. Admission control model is based on ARP-SAC incoming burst rate prediction which was published in [CT2]. 2.4 The method of reclaiming resource for high priority burst 2.4.1 The principle of reclaiming resources for high priority bursts The principle of reclaiming resources from a low priority burst for a high
priority burst is proposed as follows: When a high priority burst arrives at an output link and no wavelength is found for scheduling it, the resource which has been occupied by a low priority burst will be removed to use for scheduling this high priority burst. Low priority burst removal only is performed if 2 following conditions are satisfied: The high priority burst only overlaps with the low priority burst that is intended to remove; The control package of the low priority burst has not been sent to the next node.
Otherwise, the high priorburst is dropped. In the case that a low prior burst arrives and all resources are busy, this
burst is dropped. 2.4.2 Description of the TPAC algorithm
Algorithm 2:
Input:
- ;
//incoming burst
//number of output wavelength //the size of observation window;
- ; - ;
Output:
- Set of high QoS burst is scheduling to , and dropped ; - Set of high QoS burst is scheduling to , and dropped .
Process:
for each an burst arrives do
1
2
if ( ;) then
3
:= the current rates of high priority and low priority
4
5
;
classes arrived in the observation window; ;
6
;
7
end if
11
8
9
end for;
The complexity of the TPAC algorithm is , where n is the number of incoming bursts ( ) and is the total output wavelength. 2.4.3 TPAC analysis model
The wavelength distribution model in the TPAC is equivalent to the queue system model where there are wavelength channels (with full wavelength conversion) serving incoming bursts.
The two-dimensional Markov model [14], [28], [58] can therefore be used to analyze the efficiency (in terms of the blocking probability) for the admission control models. The simulation parameters are the same as in Section 2.2.2. Thus, the blocking probability equations of the high and low priority classes are:
(2.9)
∑
∑
(2.10)
Total probabilities of high and low priority classes is:
(2.11)
2.4.4 Comparing and evaluating based on simulation a. Comparing average prediction error
As recommended in [34], the advantage of the TW-EWMA method is that it helps to reduce computation costs. Therefore, it is essential to have a compromise between predictive error and computational cost in order to help the system achieve higher efficiency. Therefore, the thesis chooses the half size of the observation window by the predictive execution cycle (according to analysis in Figure 2.14 and Figure 2.15), and published [CT3].
b. Comparing burst loss rates Figure 2.16c shows TPAC’s burst loss rate of the both priority classes TPAC lower than all the previously proposed methods.
12
Figure 2.16 Comparison of the burst loss rate of both classes in the TPAC model
Figure 2.17 bandwidth utilization rate for both classes in TPAC model
b. Comparing of bandwidth usage
As shown in Figure 2.17c, TPAC has the highest bandwidth utilization rate for both priority classes. In addition, Figure 2.18 shows that there is an approximation between the simulation result and the blocking probability of Equation 2.11 calculated by Mathematica software [78]. This confirms the correctness of our proposed model published in [CT4].
13
Figure 2.18 Comparison on the burst blocking probability between analytical model and simulation.
2.4.5 Result 2.5 ITPAC delay line combination model 2.5.1 Description of improved iTPAC algorithm A Fiber Delay Line (FDL) control mechanism for these low priority bursts is proposed as follows:
• if the allowable delay (determined by the current offset time) of the low priority burst is less than the delay time of the FDL then it is not necessary to include the low priority burst in the delay line (because they will be dropped because not enough the offset time but the edge node has not been reached yet) and this low priority burst will be rejected; • if the allowable delay of the low priority burst is greater than the delay line length, the burst will be inserted into FDL in the hope that resources can be found to schedule when it leaves the delay line.
2.5.2 Simulating, comparing and evaluating 2.5.3 Result 2.6 Summary of chapter 2
Admission control has a large impact on the burst loss rate and the bandwidth utilization efficiency at the core node for bursts of different priority. In this chapter, the thesis has proposed an admission control model based on burst traffic prediction with 3 models ARP-SAC [CT2], TPAC [CT3], [CT4] and iTPAC [CT5].
Based on the simulation results, the models not only improve the quality of service for the high priority class but also improve the burst loss rate of the low priority class. In the next chapter, solutions to improve QoS at the edge nodes and nodes combination will be analyzed and proposed.
14
CHAPTER 3. SOLUTIONS TO IMPROVE QUALITY OF
SERVICE AT THE NODE AND COMBINE NODES 3.1 The model to differentiate quality of service at the edge node 3.1.1 The burst assembly provides quality of service
Differentiating QoS is an important requirement for actual communication networks, because the variety of service requests from users and applications is increasing. Through operations in the OBS network, QoS discrimination can be realized during burst assembly, resource allocation, scheduling, or blocking resolution. 3.1.2 Analysing of related works
There are two approaches for QoS differentiation at the edge node: offset time-based differentiation (OTD) [23], [49] and burst length-based differentiation (BLD) [25], [42]. With OTD, by adding extra offset time to the high priority bursts, these bursts are subjected to an additional delay.
However, a problem in BLD is that the edge node does not know the size of voids generated at the related core nodes, so it is necessary to feedback the value of void size from the core node so that the edge node can resize the generated bursts appropriately.
In this chapter, we will propose a model to provide QoS at the edge node and solve the problem discussed above, with the model to quality of service differentiation combining the offset time and the burst size called OT-BLD. 3.1.3 Model of providing QoS at the edge node OT-BLD
Figure 3.4 The assembly burst at the edge node of OT-BLD model The OT-BLD model is set at offset time of the low priority burst will be the basic offset time (basic OT), and the offset time of the high priority burst
The OT-BLD model burst assembly according to the following criteria: high priority burst with long offset time, but short burst length; while the low priority burst has short offset time, the burst length is long as shown in Figure 3.4.
15
Figure 3.5 Setting the additional offset time greater than the low priority burst length will
help reduce the resolution between 2 class priority burst.
is thus the sum of basic offset time and the length of low priority burst (Figure 3.5).
3.1.4 Comparison and evaluation based on simulation
Consider incoming packets at the OBS edge node queues that have Poisson's distribution and have packet sizes in the from [500, 1000] bytes. At each edge node, the number of queues is considered for each destination , consisting of a high priority queue with time threshold and a low priority queue with time threshold .
The basic offset time value is set to 300µs. The length thresholds are initially set for and , as recommended by B. Kantarci and et al in [25]. value is chosen to be , the number of wavelengths per output link is , the bandwidth of each link is .. a. Comparison of the burst loss rate between models
Figure 3.6 Comparison of the total loss rate among undiff, OTD, BLD and OT-BLD
The total burst loss rate of both classes is shown in Figure 3.6 (c), where, in the first period (from 0.0s to 1.0s, with load 0.2 for both priority classes), OT-BLD for better burst loss rates than OTD, BLD and undiff 4%, 5% and 12%.
At the second period (from 1.0s to 2.0s, when the high priority class increases to 0.4 and the 0.2 load for the low priority class), OT-BLD also
16
results in better burst loss than BLD, OTD and undiff about 3%, 7% and 10%. Thus, the OT-BLD model shows its effectiveness in terms of the increase or decrease of the incoming load. b, Comparison of the average delay of packets
Figure 3.7 Comparison of the average delay (µs) of high priority burst (packets).
The result Figure 3.7 (a) shows that the average delay of the high priority burst packets of the OT-BLD and BLD models is the lowest in both phases, while the average delay of the packets in the undiff and OTD model are higher.
3.1.5 Result
The thesis has proposed an improved model combining published burst assembly methods, with a model called OT-BLD. When comparing the results of the burst loss rate of OT-BLD, it was shown that the efficiency about burst loss of this model compared to other models. In addition, the average delay of the OT-BLD model packet decreased when compared with the undiff and OTD models in different simulation times, which also demonstrated the efficiency of the combination this model. However, the average delay of the OT-BLD model is higher than that of the BLD model, which is also the urgency for the futher work. The results presented in this section have been published in [CT6]. 3.2 Analysing the cause of the burst loss 3.2.1 Burst loss problem when scheduling The characteristics require of for the scheduling state data are: arrival
burst time, burst length, , void start and void end time.
17
3.2.2 Extract scheduling status data
NFSNET network model consists of 14 nodes. The simulation results show that the burst loss rates at the nodes 3, 5, and 8 are significant and the scheduled status data extracted here can represent the general scheduling state at the core nodes across the network. (in Figure 3.9). 3.2.3 Determining the properties that influence the burst loss
In 44959 collected samples, there were 22187 samples which belong to successful scheduling cases without voids filling, 752 successful scheduling cases with voids filling and 22019 samples which belong to unsuccessful scheduling cases. The important thing is how to know what attributes affect to the unsuccessful scheduling. The result of the scheduling data analysis in Figure 3.12 shows that head_overlap and LAUT overlap are the main causes of the burst loss (accounting for more than 90%), in which two properties burst_time and burst_length have the primary impaction on unsucessful scheduling (Figures 3.11 and 3.12). This is the basis of the following proposal to improve scheduling performance. 3.2.4 The solution to use the delay line to reduce burst loss
The scheduling control based on the FDL delay line is proposed in Figure 3.14, when an incoming burst cannot be scheduled, the LAUT_overlap and head_overlap checking are considered (the rectangular part is dashed). If there is head_overlap or LAUT_overlap, the burst will be inserted into the delay line to delay its arrival time. In the case of the tail_overlap, the burst will be dropped. 3.2.5 Simulation and analysis of the results a. Analysis of burst loss rate
Figure 315 shows that using the FDL delay line for burst loss rate from load 0.1 to 0.6 is the biggest reduction about 60% compared to no using a delay line. But when the load increases from 0.7 to 0.9, the loss rate decreases very little (about 4%). FDL with the delay of 100µs has a lower burst loss rate than FDL with a delay of 150µs.
18
Figure 3.15 Comparison of the burst loss rate when using or not using FDL
Fi
gure 3.16 The average delay (%) increases when using FDL.
b. Analysing the delay when using FDL
In Figure 3.16, the average delay rate of bursts increases when using the FDL. When using FDL with 150 µs delay, the burst delay increases on average 21% compared to nearly 16% when using FDL with 100µs delay.
3.3 Combining the edge node and the core node in the quality of
service differentiation
3.3.1 Quality of service differentiation based on offset time and adjusted burst length OT-ABLD
Chapter 3 of the thesis continues to explore the void dimension information module that is feedbacked from core nodes can be carried in the control package whose structure is described as in [44]. Specifically, void size information needs to be returned to the edge node in order for it to adjust
19
burst of generated length.
Figure 3.19 The structure of the NACK packet with 4 bytes dedicated to carry the value of void
size
This information can be carried in the NACK control package. With the structure of a NACK packet as shown in Figure 3.17, the idle bytes in the PDU CTR field are utilized to carry this value. Specifically, 4 of 6 idle bytes are proposed to carry the value of void size. The structure of PDU CTR is therefore modifield as shown in Figure 3.19.
Figure 3.20 The schema of measuring the average void size at the core node and sent it to the edge node to adjust the length threshold high priority burst.
To implement the OT-ABLD model, at the core node, the thesis uses BF- VF [45] algorithm, the best void filling algorithm to date, is chosen. Because BF-VF tests all voids and chooses the most fitting void, the average size of voids is easy to be calculated in the OT-ABLD model.
3.3.2 Comparing and evaluating based on simulation a. Comparison of the burst loss rate
Figure 3.21 shows a comparison of the total loss rate between of the QoS differentiation models: undiff, OTD, BLD, OT-ABLD. The results showed that OT-ABLD achieved the lowest total burst loss rate in both periods. Specifically, in the first period (from 0.1s to 1.0s), OT-ABLD achieves the
20
total loss rate that is 10% lower than BLD, OTD and 20% lower than undiff. In the second period (from 1.1s to 2.0s), OT-ABLD has the burst loss rate that is 15% lower than BLD, 25% lower than OTD and nearly 30% lower than undiff. 0.22
l
0.2
0.18
undiff OTD BLD OT-ABLD
0.16
s s a c 2 e t a r s s o
0.14
l t s r u B
0.12
Figure 3.21 Comparison of total loss rate among: undiff, OTD, BLD and OT-ABLD
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 Simulation time (s)
b. Comparison of the delay high priority burst (packet)
Figure 3.24 The average delay (µs) of high priority burst (packets) For OTD and undiff, although their absembly time is practically the same in both periods, but in the second period, when the density of arriving high priority burst is high (as in Figure 3.25), the length threshold La(0) always reaches first and as a result, the delay of the high priority class with to cases
For the comparison of the average delay of high priority bursts (packets), Figure 3.24 shows that OT-ABLD causes the lowest end-to-end delay when it is compared to OTD, BLD and undiff; OTD and undiff have the same average delay; and the delay tends to decrease in the second period of the simulation.
21
Void size (0.2,0.2)
Burst lenght (0.2,0.2)
50 ) s (
0
1 5 9
3 1
9 8
7 1
1 2
5 2
7 3
3 3
1 4
5 4
9 4
3 5
7 5
1 6
5 6
9 6
3 7
7 7
1 8
5 8
3 9
7 9
of incoming load (as shown in Figure 3.24). 100 e z i s d i o v r o h t g n e l t s r u B
9 2 100 successive observation windows
Figure 3.25 A comparison of the delay (μs) of HP bursts (packet) in 100 successful observation windows with two cases of incoming loads: (0.2,0.2) and (0.4,0.2)
3.3.3 Result
In the model of providing QoS at the edge and core nodes, the thesis has proposed a model called OT-ABLD, with the use of void size information to adjust the length of the high priority burst and setting up their isolated offset times. Not only increasing the efficiency for the high priority burst, the OT- ABLD model contributes to reduce the loss of the low priority burst when compared with undiff, OTD and BLD models published at [CT8] 3.3 Summary of chapter 3
In this chapter, the thesis introduces three new QoS differentiation models which are proposed: (1) the OT-BLD model to improve the QoS providing mechanism at the edge node, that is the result of the combination of OTD and BLD; (2) the model to minimize burst loss at the core node when not considering QoS; and (3) the OT-ABLD model to improve the QoS providing mechanism associated the nodes based on the providing QoS model at the OT-BLD edge node and the response from the void size from the core node.
The results of these models help to reduce significantly the burst loss rate and packet’s average delay of ther high priority class. However, these models still have limitations about the low priority burst loss rate due to not considering the adjustment in the burst assembly process for low priority class which is also a problem that needs to be overcome. In addition, it is important to add a module to calculate the average void which is sent to the edge node and periodically adjust the burst assembly process of the high priority class which make the proposed model quite complicated. The results are published in detail in [CT6], [CT7] and [CT8].
22
CONCLUSION
Optical Burst Switching on the WDM network is considered a promising technology for the next generation Internet, because OBS overcomes the technological limitations of current Optical Packet Switching and exploits flexibly bandwidth, better than Optical Channel Switching. One of the important problems in OBS network is how to enhance QoS between different service flows. For that purpose, the thesis has focused on researching models and algorithms to improve the QoS mechanism in the OBS network with different approaches. The result which are thesis has achieved include:
1. Synthesizing, analyzing, evaluating and classifying methods of improving the QoS mechanism in the OBS networks. Thereby identify the advantages and disadvantages of the algorithms and this is also the basis for proposing and improving the algorithms to improve the QoS providing mechanism at each node and both nodes.
2. Proposing 3 admission control models, which are named ARP-SAC [CT2], TPAC [CT3], [CT4] and iTPAC [CT5] to reduce the loss rate of data burst types. 3. Proposing the quality of service provding model at the edge node OT-
BLD [CT6].
4. Proposing the model to reduce loss at the core node when not considering QoS [CT7].
5. Proposal of QoS providing model which combine edge node and OT- ABLD core node [CT8] has also been proposed to optimize used bandwidth and contribute to improve the quality of service providing mechanism between service class. THESIS’S DEVELOPMENT DIRECTION From the achieved results in the thesis, the issues need to be studied in the
next time include: 1. Research the problem of improving the QoS improvement mechanism at the core node that extends more QoS layers to see the role of improving the data transmission rate and data reception rate in the network. 2. Building the new QoS providing model at the edge node, combining burst segmentation and improving QoS provision with combine the nodes with using FDL.
23
LIST OF RESULTS PULISHED BY AUTHOR
[CT1]. Pham Trung Duc. An improvement in scheduling admission control in OBS network taking into account QoS, Journal of Science and Technology, University of Science, Hue University, Jan 2018, Volum 11, Issue 1, pp. 1- 12. [CT2]. Pham Trung Duc, Vo Viet Minh Nhat, Dang Thanh Chuong. Scheduling admission control based on incoming burst rate prediction in Optical Switching Burst networks, Proceedings of the XI National FAIR Scientific Conference, 2018, pp. 137-145. [CT3]. Pham Trung Duc, Dang Thanh Chuong, Vo Viet Minh Nhat. A Model of Traffic Prediction based Admission Control in OBS Nodes, in 2019 IEEE-RIVF International Conference on Computing and Communication Technologies (RIVF), 2019, pp. 1–6, DOI: 10.1109/RIVF.2019.8713683 (SCOPUS). [CT4]. Pham Trung Duc, Le Van Hoa, Study the effect of observation window size on predictive accuracy in scheduling admission control model, Journal of Science, Hue University: Engineering and Technology (accepted). [CT5]. Pham Trung Duc, Vo Viet Minh Nhat, Dang Thanh Chuong. A scheduling admission control improvement based on the FDL incoming burst speed prediction, Proceedings of the XII FAIR National Science Conference, 2019, pp. 268-275. [CT6]. Pham Trung Duc, Dang Thanh Chuong,. QoS differentiation model based on offset time and burst size in OBS network, Hue University Science Journal: Engineering and Technology, Vol 128, No 2A (2019). DOI: http://dx.doi.org/10.26459/hueuni-jtt.v128i2A.5496. [CT7]. Pham Trung Duc, Vo Viet Minh Nhat, Dang Thanh Chuong,. OBS core node performance enhancement based on scheduling state data analysis, Journal of Information Technology and Communication, ISSN 2525-2224, Issue 02 (CS.01) 2020, pp. 53-60. [CT8]. Vo Viet Minh Nhat, Pham Trung Duc, Dang Thanh Chuong, Le Van Hoa, A mechanism of QoS differentiation based on Offset Time and Adjusted Burst Length in OBS Networks, Turk J Elec Eng & Comp Sci, ISSN 1300- 0632, Volume 28, Issue 5, 2020, pp. 2808-2820, DOI:10.3906/elk-1906-87 (SCIE).
