chapter ix high performance scheduling mechanism for ...biblio.uabcs.mx/html/libros/pdf/16/9.pdf ·...

��

Chapter IXHigh Performance Scheduling

Mechanism for Mobile Computing

Based on Self-Ranking Algorithm (SRA)

Hesham A. AliMansoura University, Egypt

Tamer Ahmed FarragMansoura University, Egypt

Copyright © 2009, IGI Global, distributing in print or electronic forms without written permission of IGI Global is prohibited.

AbstRAct

Due to the rapidly increasing of the mobile devices connected to the internet, a lot of researches are being conducted to maximize the benefit of such integration. The main objective of this paper is to enhance the performance of the scheduling mechanism of the mobile computing environment by distributing some of the responsibilities of the access point among the available attached mobile devices. To this aim we investigate a scheduling mechanism framework that comprises an algorithm provides the mobile device with the authority to evaluate itself as a resource. The proposed mechanism is based on the proposing of “self ranking algorithm (SRA)” which provides a lifetime opportunity to reach a proper solution. This mechanism depends on event-based programming approach to start its execution in a pervasive computing environment. Using such mechanism will simplify the scheduling process by grouping the mobile devices according to their self -ranking value and assign tasks to these groups. Moreover, it will maximize the benefit of the mobile devices incorporated with the already existing grid systems by us-ing their computational power as a subordinate value to the overall power of the system. Furthermore,

��

High Performance Scheduling Mechanism for Mobile Computing

we evaluate the performance of the investigated algorithm extensively, to show how it overcomes the connection stability problem of the mobile devices. Experimental results emphasized that, the proposed SRA has a great impact in reducing the total error and link utilization compared with the traditional mechanism.

IntRoductIon

Mobile computing and commerce are spreading rapidly, replacing or supplementing wired comput-ing. Moreover, the wireless infrastructure upon which mobile computing is built may reshape the entire IT field. Therefore, it is fair to say that the mobile devices have a remarkable high profile in the most common communication devices nowadays. Individuals and organizations around the world are deeply interested in using wireless communication, because of its flexibility and its unexpected and fast development. The first solution to the need for mobile computing was to make computers small enough so that they could be easily carried. First, the laptop computer was invented; later, smaller and smaller computers, such as 3G, PDAs (personal digital assistants) and other handhelds, appeared. Portable computers, from laptops to PDAs and others are called mobile devices. In recent years a great development took place on the Internet and mobile technologies. Consequently, the next step will be merging these two technologies leading to the Wireless Internet.

The Wireless Internet will be much more than just internet access from mobile devices; the Wireless Internet will be almost invisible, as people will use mobile services and applications directly. On the other hand these services and applications will be acting as our agents, conducting searches and communicating with other services and ap-plications to satisfy our needs. Not only will the integration of mobile technology and the Internet paradigm reinforce the development of the new context-aware applications, but also it will sus-tain traditional features such as user preferences, device characteristics, properties of connectivity and the state of service and usage history. Fur-thermore, the context includes features strictly related to user mobility such as user’s current geospatial location (time and/or space). As direct use of existing Internet applications in a mobile environment is usually unsatisfactory; services and applications need to take into account the specific characteristics of mobile environments. The next section will introduce an overview of mobile devices as well as the present relation model between mobile devices and the grid.

Table 1. Worldwide wireless LAN equipment shipments (1000s of units) (Navrati Saxena 2005)

��


mobile devices development

The number of individuals and organizations rely-ing on wireless devices is continually increasing. Table 1 represents a statistical study of the current and the future increase in the sales of wireless equipments and the considerable growth in the sales of mobile phones.

The mentioned table shows the rapid growth in the sales rates of wireless equipment, and they serve the purpose of being as a good metric of the flourishing future of the mobile computing. From 2001 to 2005, investments on mobile de-vices are expected to increase by 41% and reach $31 billion. In 2004, the laptops on the market reached 39.7 millions. On the other hand, not only did the number of mobile devices and wireless equipment increase, but also the computational power and the memory storage. As a result of such situation mobile computing and wireless Internet became a very important research area. This paper will approach it from the computational grid viewpoint.

mobile devices and the computational Grid

The interaction between the mobile devices and the computational grid such as depicted in figure 1 can be classified into two models:

(I) Mobile as a user of grid resources: The development in the computational power of the mobile devices such as (smart phones, PDAs... ) will be limited due to its size, battery life, bandwidth and storage of data. However, when this integration occurs all of the huge computational power and stored data of the grid will be available to the mobile client. The mobile clients send their requests to the access point (AP) which can be con-sidered as the grid gateway; the scheduler is responsible for finding a suitable resource

to perform the incoming request (Mustafa Sanver , 2004).

(II) Mobile as a grid Resource: When one mobile device is considered a resource, it will be a very inferior and low ranking resource when compared with a PC. Meanwhile, because of the large number of the mobile devices that can be used, it can be a worth computational power. Also because of its large geographical distribution it can be considered as a very excellent data collector which can be used in many applications such as (geographical information systems, weather news…). Rela-tively, there are two approaches to integrate the mobile device into the existed grid; the first is that all the information of every mobile device is recorded in the scheduler, so every device is considered to be one grid resource. The second approach is the one in which the information of the mobile devices is hidden from the scheduler, it considers all the devices connected to an access point as one grid resource and the access point responsible for scheduling tasks on the mobile devices is also connected to it.

This paper introduces “self ranking algorithm” (SRA) that will be used to build a mobile comput-

Access point

Grid

Figure 1. an overview of integration of mobile devices with computational grid

��0


ing scheduling mechanism. Before introducing the proposed algorithm, an overview about the related work in the scheduling mechanism in the grid is given in section 2, detailed description of the targeted problem will be presented at hand in section 3, and a proposed framework is explained in section 4. Moreover, the proposed SRA will be introduced in section 5, the simulation used to state the proposed algorithm is available in section 6, finally, the results of that simulation will be analyzed in section 7.

RelAted WoRk

Before starting to elaborate on the problem, five of the most recent systems especially on the sched-uling algorisms are studied (Xiaoshan He 2003; Rajkumar Buyya 2003; Arun A. Somasundara 2004; F. Berman; H. Casanova 2005). Although they have very different parameters and concepts, all of them have two main objectives. The first objective is to increase the utilization of the system, while the second one is to find a suitable resource (as the economic cost, QoS, deadline…).

Table 2 shows a comparison between the most recent systems. Undoubtedly, one of the common problems that face any system when dealing with a large number of resources is “Load Balancing”. Due to the fact that the ranking value of the resources is different, each of these systems endeavors to solve the problem which is illustrated in table 2. Another problem is how the system will deal with the mobility of clients and resources. The noticed remark is the limitation of researches that take into account the mobility of the resources (Daniel Nurmi 2004 ; Sang-Min Park 2003) .The study of these five systems shows that they are based on different parameters to rank the resource, but the most popular are QoS and the economic cost (Xiaoshan He 2003 ; Rajkumar Buyya 2003). The expressions used in Table 2 will be explained underneath:

Backfilling: it is a technique that tries to fill the gaps in the scheduling operation by execut-ing the low priority functions in the low ranking resources that have not been used for a long time. This increases the system overall utilization and makes a kind of load balancing between the re-sources (Arun A. Somasundara , 2004).

Resource usage Accounts (Quotas): each resource must be assigned to certain functions according to its usage account. So, that preventing the resource from not being used can be caused by of the presence of high QoS resources. This approach gives the scheduler force more functions to be assigned to a certain resource by maximiz-ing its Quota (Jang-uk In , 2004) .

Job Proxy: is created when the mobile user submits a job, it is responsible for the interaction between the mobile device and the system. It can also simulate the mobile action in case of mobile disconnecting. It does this until the mobile is connected again. If the mission is accomplished and the mobile is still disconnected, it stores the result for certain time-out duration (Sang-Min Park, 2003).

QoS Guided: the QoS Guided scheduler is the kind of scheduler that has a kind of intelligence as not to consume the high QoS resource in per-forming the jobs that need low QoS. It does this to save its power to the other tasks that need this high QoS (Xiaoshan He ,2003) .

schedulInG And connectIon stAbIlIty PRoblem

The new approach in the computing area is the Internet computing. It uses the already existing infrastructure of the Internet, and builds its own grid using devices interconnected to the Inter-net (Frontier, 2004). This is a very economical approach, because of the needless of building a special infrastructure. On the other hand, a lot of questions and issues raise, such as: “Do we

��


need to build a new infrastructure of a grid to integrate the mobile devices as a grid user or as a grid resource?” and also “What about the already existing grid projects?” (Peter Gradwell, 2003; Holly Dail, 2002; James Frey, 2001; F. Berman, 2005). Figure 2 shows how the already existing infrastructure can be ordered and organized to create an infrastructure that helps to integrate the mobile devices with the existing grid systems like (Condor, GriPhyN, Grid2003). This infrastructure aims at using the huge computational power due to the large number of internet users. It also aims at using the different services and resources avail-able in the already existing grid projects. Above all, the main objective is to use the internet net-work to connect the mobile devices to the other parts of this infrastructure and to put all these services and computational power available to the mobile device. Finally, it aims at increasing the computational power and number of services of the system by integrating all that large number of mobile devices distributed around the world (Navrati Saxena , 2005).

The most important problem that can face any grid system is to develop a scheduling mechanism to manage such integration. The previous schedul-ing mechanisms depended on QoS (Xiaoshan He, 2003), cost (Rajkumar Buyya, 2002; Alexander Barmouta, 2003) or hybrid between some of other parameters (Jang-uk In, 2004; Atsuko Takefusa, 2001) to select the best scheduling decision. Due to this integration and the mobility of the device, a new parameter appeared. This parameter rep-resents the stability of the connection established between the devices and the access point, in other words the rate of disconnecting and the rate of reconnecting. All the already existing systems make the scheduler monitor and evaluate the performance and the availability of its attached resources. This was acceptable with the PCs, but because of the huge number of mobile devices expected to attach to the scheduler, a very high overload on the scheduler can happen. So, the scheduler slows down more and more as the number of the attached resources increases.

Table 2. Comparison between Referenced Systems

��


Plan of solution

To overcome the overhead resulted from collect-ing the data at the access point scheduler and storing the historical ones of the mobile device performance, a SRA will be investigated. This algorithm has two key points, the first is to provide the mobile device with authority to evaluate and rank itself and remove this task from the central point (scheduler). The second one is considering the mobility of the resource as important metrics in such environment. Therefore, the main aim of this algorithm is to calculate a ranking value for each attached mobile device, which may be considered as a metric of the mobile performance. Moreover, it will be used to classify the mobile devices into groups to make the process of sched-uling simpler and faster.

PRoPosed fRAmeWoRk

Figure 3 depicts the framework and system com-ponents relationship for the given organization

in figure 2. The following design guidelines are required to be adhered: (I) Use opportunistic schedulers which introduced in the Condor (Arun A. Somasundara , 2004), because it is a very excellent idea to make a good load balancing between high ranking resources and low ranking ones (e.g. Mobile devices). (II) Use the mobile proxy which is introduced in (Sang-Min Park, 2003) but we changed its name from job proxy to our proposed name “mobile proxy” which will be the interface between the mobile client and the other components of the system. (III) Use multi Schedulers because of the distribution of the considered infrastructure.

Proposed framework entities

In the following, the entities participating in the given framework are defined, their functions, and how they interact with each other as well.

The Task Farming Engine (TFE): it is respon-sible for partitioning the requested job into small tasks which will be assigned to resources to per-form them using the scheduler and the dispatcher.

Figure 2. the System Infrastructure organization

APAP

AP

Exist Grid

Exist Grid

R1

R2 Ri: Grid resource

��


The Scheduler: it is responsible for resource dis-covery, resource trading, resource selection, and tasks assignment. The Dispatcher: it is responsible for the actual assigning of tasks to the resources decided by the scheduler, monitoring the execution of the tasks and controlling the process of collect-ing the different partitions of the job. Finally, it sends the overall result to the job requester. Grid information System (GIS): it can be considered as the resources characteristics database used by scheduler to find a suitable resource to perform the requested tasks using the resource QoS, Cost, rank. Dedicated Scheduler: each resource is as-signed to one dedicated scheduler who has all the rights to use the resource at any time except if the resource owner needs his resource. This monopoly may leads to dis-functioning of some resources because they are in the resources list of certain Dedicated Scheduler beside other high ranked resources. So, these high ranked resources

will be preferred to the scheduler. This problem may be resolved by the temporary claiming of the resource to other type of scheduler named “opportunistic scheduler”. This problem causes holes in the scheduling operation. Opportunistic scheduler: when the dedicated scheduler claims some of its resources because they were idle for a long time or they had a low ranking value which made them useless for a long time. The Opportunistic scheduler tries to use this resource to execute some small tasks that may end before the dedicated scheduler needs the resource again. This operation is named “Backfilling”, it is noted that this method will maximize the utilization of overall system.

Now, if a mobile client is connected to an ac-cess point, the first step is to create a mobile proxy object, which will be considered as a simulation of the mobile device. So, it may store the hardware specification of the mobile and its current loca-

Figure 3. Mobile Dev. Scheduling Framework and components relationship

��


tion and it may also monitor the movement of the mobile from access point to another. This mobile proxy information will be the base knowledge on which the scheduler builds its work

Figure 4 depicts a request processing scenario. If the mobile client makes a request, this request will be stored in the mobile proxy. Then, it goes to a scheduler using the scheduling mechanism trying to find a suitable resource to perform this request from its local connected resources. But if the access point scheduler does not find a suitable resource it forwards this request to a higher level scheduler which usually has static PCs which have more computational power. This scheduler uses the GIS to find a suitable resource, when the resource is located the dispatched assigns the requested task to this resource. When the task is performed the outcome returns to the mobile proxy which is responsible for sending this result to the mobile clients in their current location.

PRoPosed “self RAnkInG AlGoRIthm” sRA

The idea of the “Self ranking Algorithm” is to reduce the dependability on the access point scheduler and to distribute this overhead among the attached mobile devices. This can be done by making every mobile able to evaluate itself. Then, the access point can use this ranking value in the process of the scheduling.

The triggering to start this algorithm execu-tion depended on the event based programming approach. The events that were taken into account are, (I) the event of disconnecting the mobile de-vice and its scheduler because this event means the end of the last connected period, (II) the event of reconnecting the mobile device to its sched-uler because this event means the end of the last disconnected period, (III) the event of finishing a task because this event changes the value of

Figure 4. Request processing flow

Search in its local resources

Mobile Request

Mobile Proxy

Access point Scheduler

Grid Scheduler

Find a resource

Dispatcher

Search in GIS for a resource

Yes

No

Perform the task and return the

result

Return false

Yes

No

Perform the task and return the

Find a resource

��


the mobile utilization. The self ranking value (R) has two parts: the first is the Connectivity met-ric (MCD) which can be considered as a metric of performance and connectivity of the mobile device as well. While, the second part which is the utilization metric (U) can be considered as a metric of the success of the mobile device in performing the assigned task. When the mobile client has a new ranking value, this value must be sent to the mobile proxy to be entered as a parameter in the scheduling process.

The considered parameters which are be used in the SRA are: The average number of time units being connected continually (Caverage), the average number of time units being discon-nected continually (Daverage), and the previous utilization history metric (U). The calculated values of Caverage and Daverage will be used as a key to the proposed ranking map which is used to calculate the first part of the rank value that measures the mobile performance and con-nectivity. The overall ranking value is assumed to be between 1 and 10. This part represents 80% of this value; this percentage can be changed ac-cording to the schedulers’ administrators. Figure 5 shows the rank metric map which is based on two roles, first as the Caverage value increases,

the rank must increase also. Second Daverage value increases the rank must decrease. The Cav-erage and Daverage is used to calculate values. It works as a coordinator of the Connectivity metric (MCD) on the rank metric map. The second part of the ranking value is the metric of the utiliza-tion of the mobile devices. So, it is calculated by the ratio between the number of the successful tasks and the number of all tasks. Summation of the two parts will generate the overall ranking value of the mobile device. Figure 6 shows the proposed algorithm.

sImulAtIon model

Validation of the proposed algorithm is done via simulation. The investigated simulation program is composed of three modules. The first one is responsible for generating a random movement path for the mobile devices, while the second is responsible for tracking the generated path, and this will be done through the access point. Finally, the third is responsible for tuning of critical parameters values and collecting outputs parameters, which are required to calculate Cav-erage and Daverage.

Figure 5. Rank metric map

Units average

DisconnUnits average

Low

High

Average

High Low Average

8

0

7

4 5

1 2

6

3

Caverage

Daverage

MCD is 5

��


mobile device movement mechanism

The mobile device movement path that will be generated is based on a mechanism that guarantees a random path as follows:

1. Generate random black and white areas as shown in Figure 7-a, white areas imply that there is an available connection between the mobile device and the scheduler access point, on the other hand black areas depict

Figure 6. Self Ranking Algorithm for determine the mobile device ranking value

0 Imports the prestored values :ts : the start of the last reconnection event. te: the end of the connection period .

Tu :Time unit defined by the scheduler Nc : number of reconnecting event occurring.

ND: number of disconnecting event occurring.

Caverage: The average number of time units to being connecting continually

Daverage : the average number of time units being disconnecting continually

Ns : the number of successfully performed tasks N : total number of tasks assigned to the device.

1 Wait for incoming event and check it.2 if the event is Disconnecting Event at time ( t ) then :

2.1 Replace the prestored value te with the new value t : te = t

2.2 Calculate the Connection Period Pc by using the stored value of ts and te: Pc = te –ts2.3 Calculate number of time units Xc of the connection period Pc by using Tu provided by the

Scheduler :2.4 Calculate the new value of Caverage by using the prestored value of Caverage and the prestored

Nc:

1/NXNCC ccc*(old)averagew)average(ne 2.5 Nc = Nc+1.

2.6 Calculate the connectivity metric MCD by using the new calculated Caverage and the prestored

Daverage as coordinates of a point in the "rank metric map".

3 if the event is reconnecting Event at time ( t ) then :

3.1 Replace the prestored value ts with the new value t :ts = t

3.2 Calculate the disconnection Period PD by using the stored value of ts and te: PD = ts –te3.3 Calculate number of time units XD of the disconnection period PD by using Tu provided by the

Scheduler :

uDD TPX /3.4 Calculate the new value of Daverage by using the prestored value of Daverage and the

prestored ND:

1/NXNDD DDD*(old)averagew)average(ne 3.5 ND = ND+1.

3.6 Calculate the connectivity metric MCD by using the new calculated Daverage and the prestored

��


disconnection.2. Divide the whole area into small rectangular

areas as shown in Figure 7-b.3. Generate a point within each rectangle at

random position as shown in Figure 7-c.4. Save the position of the generated points in

an array.5. Select one point from the previous array

in random fashion to be the starting point

of the movement path as shown in Figure 7-d.

6. Select one of the possible eight directions shown in Figure 8 for the next hop.

7. Continue the movement towards the previous selected direction for a random number of hops.

8. Repeat step 6 and 7 until having the required length of the movement path.

9. Store all the selected points in step 6, 7 and 8 to represent a path for a mobile device movement.

10. Repeat steps from 2-9 to generate another mobile movement path.

Figure 9 illustrates some examples of the generated random mobile movement paths based on the previous mechanism.

Figure 7. steps of Random movement path generation

(c) (b) (a) (d)

Start point

Figure 8. choose a random direction from eight

Start point

1

2

5

4 3

8 7 6

Figure 9. Examples of the random mobile movement paths

��


AP and monitoring the mobile device

This module, as stated before, simulates the AP monitoring of the mobile device movement pro-cess. In such process AP sends “Are you alive?” message. If there is an available connection, the mobile device responses by an “I’m alive” mes-sage. The time between sending and receiving is called the response time Tr, this time can be determined experimentally, the AP waits for an-other threshold time Tth before sending the next monitoring message . On the other hand, if there is no response for Tr, the access point will send a message again. According to the response of the previous simulation the AP reports the mobile device status. Figure 10 shows this process.

At this point we have to notice that reduc-ing Tth will lead to more accurate results, but on the other hand, the number of messages will increase which means high link usage which is considered from the application point of view to be a bad usage.

Parameters setting, collecting and calculating

The different parameters, which are required for comparing the self ranking against the traditional AP ranking from network utilization and accuracy point of view, are calculated in this module.

Firstly, the speed of the mobile device move-ment and Tth and Tr is tuned. Some parameters

Figure 10. AP monitoring of the mobile device movement process

Mobile Device

Access Point

Tr

Tth

(a) "Are you alive" with response

(b) "Are you alive" with no response

"Are you alive" Message "I'm alive" Message

Mobile Device

Tr Tth Tr Access Point

��


Figure 11. Total error and link utilization at no of mobiles = 50

tth = 2

tth = 4

tth = 6

0.00

5.00

10.00

15.00

20.00

25.00

30.00

0.00 10.00 20.00 30.00 40.00

mobile movement speed (m/s)

tota

l err

or

speed 30 m/s

020000400006000080000

100000120000140000160000180000200000

2 4 6value of tth (seconds)

nu

mb

er m

essa

ges

no. of messages Using Self Monitoring

no. of messages Using AP Monitoring


tth = 2

tth = 4tth = 6

0.00

10.00

20.00

30.00

40.00

50.00

0.00 10.00 20.00 30.00 40.00


tota

l err

or

speed 30 m/s

0

100000

200000

300000

400000

500000

600000


nu

mb

er m

essa

ges no. of messages Using Self Monitoring


tth = 2

tth = 4tth = 6

0.005.00

10.0015.0020.0025.0030.0035.00

0.00 10.00 20.00 30.00 40.00


tota

l err

or

speed 30 m/s

0

50000100000

150000

200000

250000300000

350000

400000


nu

mb

er m

essa

ges no. of messages Using Self Monitoring



��0


from the first and second modules are collected and stored including: the length of the generated path, the number of connection and the discon-nection during the movement on the path, the total number of “Are you alive?” messages, and the number of messages with and without response. So, Caverage and Daverage can be calculated.

PeRfoRmAnce AnAlysIs And dIscussIon

Based on the previous discussion, on the change of the number of mobile devices used during the experiment (50, 75 and 100 mobiles) or on the change of the value of Tth (2, 4 and 6 seconds) various experiments are performed. There are another two factors were constant; the first is the length of the movement path witch selected to be long relatively (10000 hop). The second is Tr which is selected to be small relatively (0.5 second). Each of these experiments will be repeated for different movement speed from low mobility (with average movement speed 2 m/s) to high mobility (with average movement speed 30 m/s).

The average error in calculating the Caver-age and Daverage has been calculated for each experiment at each used speed and their summa-tion represents the total error in the experiment. Also, the number of network messages exchanged between AP and the mobile device, in both cases the AP monitoring and self monitoring, has been counted.

Figures 11, 12 and 13, show that the percent-age of the total error increases rapidly as the movement speed of mobile device increases. This result is expected because as the movement speed increases the ability of AP to sense the change in the mobile connectivity will be more and more limited. Also, the figures show that when the value of Tth increases the percentage of the total error increases also while the number of the exchanged network messages decreases. This result is ex-pected because Tth represents the time between

two monitoring messages, as this time increases that means reduction in the ability of AP to sense the change in the mobile connectivity. The fig-ures show the comparison between the number of exchanged messages between AP and mobile devices in the case of self monitoring and case of AP monitoring. It is noted that, in case of AP monitoring the number approximately doubled more than 70 times compared to the case of self monitoring.

conclusIon

This paper points out an overview of the issues of mobile devices integration with the existing grid. It shows that when some of authorization is impeded within the mobile client, every mobile can evaluate its performance by itself. The tradi-tional method makes an overhead on the scheduler to perform a historical evaluation to the mobile performance which makes it busy in a secondary task and leaves its main task of scheduling. So, the “self ranking algorithm” will be the base of a scheduling mechanism which will schedule the tasks on the mobile devices. The originality of the proposed mechanism concentrates on mobile cooperating with services at the access point (AP). Using such mechanism will lead to minimizing the calculation time consumed in mobile rank-ing and evaluating before starting the scheduling process. Moreover, it will lead also to minimizing the amount of the stored data at the scheduler and to simplifying the scheduling process by grouping the mobile devices according to their self ranking value and assign tasks to these groups. Finally, it will result in maximizing the profit of the mobile devices integrated with the already existing grid systems by using their computational power as an addition to the system overall power. In brief the outcome will be maximizing the system utilization and making the system more flexible to integrate any new devices without any affright to increase the system complexity.

��


In this paper we present the newly emerging technical issues for realizing this mobile grid sys-tem, and particularly focus on the job scheduling algorithm to achieve more reliable performance. However, there are still challenging problems such as limited energy, device heterogeneity, security, and so on. We will tackle on these is-sues in future works and develop a prototype of mobile grid system.

RefeRences

Alexander Barmouta & Rajkumar Buyya (2003), “GridBank: A Grid Accounting Services Archi-tecture (GASA) for Distributed Systems Shar-ing and Integration “, Parallel and Distributed Processing Symposium, Nice, France, IEEE Computer Society.

Anthony Sulistio & Chee Shin Yeo & and Raj-kumar Buyya (2003), Visual Modeler for Grid Modeling and Simulation (GridSim) Toolkit, International Conference on Computational Science (ICCS) 2003: Saint Petersburg, Russia / Melbourne, Australi .

Arun A. Somasundara & Aditya Ramamoorthy & Mani B. Srivastava (2004), Mobile Element Scheduling for Efficient Data Collection in Wire-less Sensor Networks with Dynamic Deadlines , 25th IEEE International Real-Time Systems Symposium (RTSS’04) .

Atsuko Takefusa (2001) , A Study of Deadline Scheduling for Client-Server Systems on the Computational Grid, 10th IEEE Symposium on High Performance and Distributed Computing (HPDC’01), San Francisco, California.

Daniel Nurmi & Rich Wolski & John Brevik (2004), ModelBased Checkpoint Scheduling for Volatile Resource Environments, University of California Santa Barbara, Department of Comput-er Science, Technical Report, Santa Barbara.

Dantong Yu (2003), Divisible Load Scheduling for Grid Computing, Conference on Parallel and Distributed Computing and Systems (PDCS 2003).

F. Berman & H. Casanova (2005), “New Grid Scheduling and Rescheduling Methods in the GrADS Project”, International Journal of Parallel Programming.

Frontier (2004), The Premier Internet Comput-ing Platform, Whitepaper, http://www.parabon.com/clients/clientWhitePapers.jsp.

Holly Dail & Henri Casanova & Fran Berman (2002), A Decoupled Scheduling Approach for the GrADS Program Development Environment, IEEE/ACM SC2002 Conference.

James Frey (2001), Condor-G: A Computation Management Agent for Multi-Institutional Grids, IEEE.

Jang-uk In & Paul Avery (2004), “Policy Based Scheduling for Simple Quality of Service in Grid Computing”, 18th International Parallel and Distributed Processing Symposium (IPDPS 2004),Santa Fe, New Mexico, USA. IEEE Com-puter Society.

Mustafa Sanver & Sathya Priya Durairaju & Ajay Gupta (2004), “Should one incorporate Mobile-ware in Parallel and Distributed Computation?”, 10th International Conference on High Perfor-mance Computing (HiPC 2003) , Hyderabad, India.

Navrati Saxena & Kalyan Basu & Sajal K. Das & Christina M. Pinotti (2005), “A Dynamic Hybrid Scheduling Algorithm with Clients’ Departure for Impatient Clients in Heterogeneous Environ-ments” , 19th IEEE International Parallel and Distributed Processing Symposium (IPDPS’05), April - Rhodes Island, Greece.

Navrati Saxena (2005) , New Hybrid Scheduling Framework for Asymmetric Wireless Environ-ments with Request Repetition , Third Interna-

��


tional Symposium on Modeling and Optimization in Mobile, Ad Hoc, and Wireless Networks.

Peter Gradwell (2003) ,Overview of Grid Schedul-ing Systems, Located at http://www.peter.me.uk/phd/writings/computing-economy-review.pdf

Rajkumar Buyya & Manzur Murshed (2002), GridSim: a toolkit for the modeling and simula-tion of distributed resource management and scheduling for Grid computing, The Computing Research Repository (CoRR).

Rajkumar Buyya & David Abramson & Jonathan Giddy (2002), Economic Models for Resource Management and Scheduling in Grid Computing, The Journal of Concurrency and Computation: Practice and Experience (CCPE).

Rajkumar Buyya & Manzur Murshed & and David Abramson (2003), A Deadline and Budget Constrained Cost-Time Optimization Algorithm for Scheduling Task Farming Applications on Global Grids, The Computing Research Reposi-tory (CoRR).

Sang-Min Park & Young-Bae Ko & Jai-Hoon Kim (2003) , Disconnected Operation Service in Mobile Grid Computing, International Con-ference on Service Oriented Computing ICSOC 2003: Trento, Italy.

Xiaoshan He (2003), A QoS Guided Scheduling Algorithm for Grid Computing, Journal of Com-puter Science and Technology, Special Issue on Grid Computing, 18(4).

This work was previously published in Int. Journal of Information Technology and Web Engineering, Vol 1, Issue 2, edited by G. Alkhatib and D. Rine, pp. 43-59, copyright 2006 by IGI Publishing (an imprint of IGI Global).

chapter ix high performance scheduling mechanism for ...biblio.uabcs.mx/html/libros/pdf/16/9.pdf ·...

Documents