Medf: Deadline Driven Algorithm For Minimizing Response Time and Completion Time in Iaas Clouds

International Journal of Application or Innovation in Engineering& Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org

Volume 3, Issue 6, June 2014 ISSN 2319 - 4847

Volume 3, Issue 6, June 2014 Page 16

Abstract
Cloud Computing is used to avail internet based services on a lease basis in a very short time period. Purchase and installation of
new computing resources takes time, so getting resources as a service from cloud in a minute or so is faster and cheaper
nowadays. A contract called service level agreement is worked between cloud host and consumer. SLA includes quality of service
(QoS) that must be provided by the cloud host to consumer. It also includes time frame, in that time period consumers task should
get done. Due to multi-tenancy architecture a cloud host has many consumers at same time. Some consumers cannot get their job
done in time limit. When rejecting consumers leases, deadline plays an important role. This paper introduces mEDF (Modified
Earliest Deadline First) algorithm and leasing policy for deadline driven resource management, which takes care of minimum
capacity expenses at cloud host side. Experiments and results show that this algorithm is suitable and can be used for deadline
driven scheduling with minimizing response time and completion time.

Keywords: Deadline, Resource management, IaaS cloud, mEDF.

1. INTRODUCTION
Cloud computing is the practice of using a network of remote servers hosted on the Internet to store, manage, and process
data, rather than a local server or a personal computer [1]. These remote servers show infinite capacity. Resources
provided by cloud can be increased or decreased on demand. Consumers have to pay only for what they have used. Proper
monitoring of SLA is done to provide QoS to consumers.
Clouds can be public, private or hybrid. In a public cloud many consumers from same or different organization can be
served at a time. In a private cloud, consumers only from the same organization are served from the cloud. Hybrid model
of cloud allows some services from its own resources and some from other cloud providers. Cloud Computing provides
three different types of services namely (1) Software as a Service (SaaS), (2) Platform as a Service (PaaS), and (3)
Infrastructure as a Service (IaaS).
In SaaS model of cloud, QoS-based workflow scheduling algorithm based on Partial Critical Paths (PCP) can be used [2].
PCP tries to minimize the cost of workflow execution while meeting a user-defined deadline. Recursively scheduling the
partial critical paths ending at previously scheduled tasks can also be done.
User defined deadline can also be provided in IaaS clouds. It may be a situation that a consumers task is to run in an
unattended mode with a predefined deadline that either provide resources in a given time frame or reject the request.
Haizea, which can be used as a plug-in scheduler for opennebula provides deadline driven scheduling [3]. This deadline
driven policy does not account for deadline sensitivity with minimizing computing capacity requirements to cloud host.
Our proposed policy and algorithm minimizes computing capacity requirements while keeping earliest deadline meeting
on the priority. Algorithm to minimize computing capacity requirement in deadline driven scheduling for different
consumers tasks are proposed in this paper. Experimental results are also shown to advocate the algorithm.
Rest of the paper is organized as: Section 2 contains related work. Section 3 describes need of introducing mEDF
algorithm for completing consumers task. In section 4, proposed algorithm, mEDF algorithm to minimize computing
capacity while meeting deadlines and implement better resource allocation policies, are given. Experimental results of
algorithm are presented in section 5. Section 6 contains conclusions and future work.

Figure 1 Consumers request, Revenue generation and deadline are interrelated to each other.
mEDF: Deadline Driven Algorithm for
Minimizing Response Time and Completion
Time in IaaS Clouds

Vivek Shrivastava
1
, and D. S. Bhilare
2

1
International Institute of Professional Studies, Devi Ahilya University, Indore, India
2
Computer Centre, Devi Ahilya University, Indore, India


2. LITERATURE SURVEY
Cloud Computing offers leasing computing capacity on a pay per use basis based on service level agreement. It saves time
and cost of installing hardware and software at consumers premises. Cloud provides services which are elastic in nature.
[4], [5].
Computing infrastructure is provided to consumer in the formof virtual machines that can be used for different types of
needs. IaaS service model provides tools and technologies in such a way that an organizations existing hardware can be
used to provide hardware services to other organizations or same organization. So an organization can use a cloud for its
services or can become a cloud host and provide services to others [6].
In [7] provides an algorithmand a leasing policy for prioritizing consumers on the basis of CRI score. Authors showed
policy to maintain order of execution of users task on the basis of some parameters and claimed their result is an
improvement over existing policies and algorithms under some conditions.
Improvement on automated allocation and de-allocation of resources, resource monitoring, resource discovery and
registration are provided in [8]. This work increases chance of adoption of clouds by the consumers. [8], [9].
Anekas deadline-driven provisioning mechanism is given in [10]. This mechanismis responsible for supporting quality of
service (QoS)-aware execution of scientific applications in hybrid clouds. This mechanismcan use and manage resources
obtained from same or a variety of sources that include desktop grids, clusters, grids, public and private clouds.
Particle Swarm Optimization (PSO) is utilized to schedule workflows in the algorithm proposed in [11]. This was
developed to work in clouds with a single-level SLAs and on-demand resource leasing. It does not consider workflow
deadlines, but focuses only on monetary cost minimization.
IaaS cloud can be setup by using toolkits OpenNebula [12], Eucalyptus [13], OpenStack [14] etc.. Currently, OpenNebula
scheduler provides an implementation of the Rank Scheduling Policy which aims to prioritize those resources more suitable
for the VM. Match making scheduler of OpenNebula works on the match making algorithm. This algorithmworks on policy
from0 to 4. 0th policy is packing, which Minimize the number of hosts in use by packing the VMs in the hosts to reduce
VM fragmentation, 1st policy is stripping, which Maximize resources available for the VMs by spreading the VMs in the
hosts policies provided. 2nd policy is load-aware that maximize resources available for the VMs by using those nodes with
less load. 3rd policy is Customthat uses a custom RANK. 4th policy is named Fixed in which, hosts will be ranked
according to the PRIORITY attribute found in the Host or Cluster template.
Haizea, a toolkit developed in Python to work as a customscheduler for opennebula or in simulation mode was introduced
in [15]. New tailor made policies can be developed in Haizea. Haizea basically provides four types of leases: Best-effort
(BE), Advanced reservation, Immediate leases and Deadline sensitive leases [16]. BE leases are preemptable and breakable
i.e. resources can be given in different time intervals to every BE leases. BE leases does not have any time constraints, these
leases are served as and when lease load permits the system. Immediate and advance reservation (AR) leases are non-
preemptable and have time constraints, such as start time and end time. AR leases have more priority than BE leases for
resources. It will preempt best-effort leases whenever the resources are required for advance reservation. There is no
guarantee that a submitted best-effort lease will get resources for completion within a certain time limit [17].
If the systemhas lots of AR and immediate leases then priority will be given to these leases. BE leases have to wait for a
long time and will be served when resources are free fromAR & immediate leases [18]. To protect these leases from
starvation, Starvation-removal algorithmwas proposed in [19]. This algorithmsupports very less suspensions of leases by
counting number of suspensions and then increases priority of that lease by aging mechanism.
BE leases takes long period of time to complete due to presence of so many AR and immediate leases. This may lead to
more requests to AR leases which in turn may cause internal fragmentation of free resources. Deadline sensitive (DLS)
leases can also be used with Haizea. DLS are assumed to be preemptable but there is a limitation to their preemptability. It
is preemptable only if the scheduling algorithmof Haizea can assure that it can be completed before its deadline. Thus, it
will assure (1) to consumers that their request will be completed within a certain time limit (2) minimuminternal
fragmentation [20], [21], [22].
In [16], authors extended deadline sensitive algorithm of Haizea by considering several time slots that together can satisfy
complete lease within given deadline. Leases with same resource request or if first lease request has fewer resources then
second lease then only swapping of lease can be done to handle deadline situation. Other method proposed by the authors
was backfilling but unfortunately, they did not implement it and left it as future work.
In [24] authors proposed Auto-scaling approach to minimize cost and meet application deadlines in cloud workflows. In this
approach, the basic computing elements are virtual machines (VMs) of various sizes/costs, jobs. These VMs jobs are
specified as workflows. Users specify performance requirements by assigning (soft) deadlines to jobs, and the goal is to
ensure all jobs are finished within their deadlines at minimumfinancial cost. Authors proved their work by dynamically
allocating/deallocating VMs and scheduling tasks on the most cost-efficient instances.
The deadline-driven cost-minimization algorithm [25], or the Deadline-Markov Decision Process (MDP), breaks the
workflow into partitions. It assigns a maximumfinishing time for each partition according to the deadline set by the user.
Based on this time, each partition is scheduled for that resource which will result in the lowest cost and earliest estimated
finishing time. This algorithm works with on-demand resource reservation.


Deadline and Budget distribution-based Cost-Time Optimization (DBD-CTO) workflow scheduling algorithm was
proposed in [26]. This algorithm minimizes execution cost while meeting timeframe for delivering results, but this
algorithms work with real time workflows left as future work. In our work, we are applying earliest deadline first
algorithm[27] with a modification that, it should minimize the capacity required at cloud host side.

3. NEED OF MEDF ALGORITHM
Earliest deadline first (EDF) or least time to go is a dynamic scheduling algorithm. EDF is mostly used in real-time
operating systems to place processes in a priority queue. Processes are arranged in queue according to their deadlines.
Process queue is searched on every event of existing process finishes or a new process arrives for the process which is
closest to its deadline. This process is the next to be scheduled for execution [28].
In IaaS cloud environment, lease requests arrivals are independent. These lease requests can be characterized by arrival
time, duration and a deadline. Number of nodes required by the lease request can also be a factor that can play a vital role
for optimizing computing capacity requirement at cloud host side in scheduling. Our mEDF algorithm can take these factors
into account and can lead to improve waiting time and completion time of lease requests.

4. PROPOSED ALGORITHM
Algorithmrequired to implement mEDF policy are given in this section. Computing environment, test lease requests and
algorithmare described in next subsections.
Computing Environment
A collection of 4 nodes, all with one CPU and 1024MB of memory are used as system for testing in simulated mode of
Haizea tool kit.

<nodes>
<node-set numnodes="4">
<res type="CPU" amount="100"/>
<res type="Memory" amount="1024"/>
</node-set>
</nodes>

Sample Lease request

<lease-workload name="sample">
<description>
A simple trace where only one lease with deadline is specified is shown as sample here.
</description>
<lease-requests>

<lease-request arrival="00:00:00">
<lease preemptible="true">
<nodes>
<node-set numnodes="2">
<res type="CPU" amount="100"/>
<res type="Memory" amount="1024"/>
</node-set>
</nodes>
<start></start>
<duration time="10:00:00"/>
<deadline>
<exact time="01-01-2014 00:00:00"/>
</deadline>
<software>
<disk-image id="foobar.img" size="1024"/>
</software>
</lease>

</lease-requests>


Attributes and their sub attributes like lease, lease id, nodes, start, duration time, and software are already available with
Haizea installation. We have introduced new attribute deadline, so that a consumers tasks can be given a deadline. Monte
Carlo simulation is used for generating durations and Poisson distribution is used for lease request arrival times for
experiments.
Haizea may use backfilling algorithmwith aggressive, conservative and off settings. We have used default aggressive
settings for our experiments. Rejection of leases which have crossed deadline limit has also been developed. Lease requests
arrive in their natural order. mEDF sorts all the lease requests in increasing order of deadline and number of nodes required
for leases and enqueue themtime to time taking care of backfilling process.

Algorithm
Start
For all leases i =1 to n do
Begin
Enqueue request as arrived in queue
Sort leases according to Deadline, number of number of nodes
If Numnodes_requested <=Numnodes_available and deadline >=available_schedule then
Accept lease
Else
Begin
Reject lease
If deadline >=available_schedule then
Reschedule lease
Else
Delete lease entry fromqueue
End
If lease completed then
Delete lease entry fromqueue
End
Stop

5. EXPERIMENTS AND RESULTS
Table-1 and Table-2 shows consumers lease requests under test run with existing DLS and mEDF algorithmrespectively.
This batch of consumers lease requests is generated only for checking the validity of our proposed leasing policy and
algorithmin lab. Parameters Lid, Req. Arrival, Duration in days, Deadline, NumNodes, Response time are used in table,
where Lid is lease id usually given in ascending order in order of appearance of leases, Req. Arrival is date and time of
consumers lease request, Duration in days is also a parameter which specifies requested amount of time for cloud services ,
NumNodes is number of nodes required by consumer in present lease, and Response time is time at which first response
given to a particular lease and completion time is last date and time of providing nodes to consumers lease request.
Graph in figure-2 shows comparison between DLS algorithmand mEDF algorithmtested in Haizea tool kit.

Table -1: Deadline sensitive algorithm
Lid Req. Arrival
Duration
in days Deadline
Num
Nodes
Response
time Completion time
1
01-01-2014 12:00
AM 10
14-01-2014
12:00 AM 2 01-01-2014 10-01-2014
2
01-01-2014 12:00
AM 8
12-01-2014
12:00 AM 2 01-01-2014 08-01-2014
3
01-01-2014 01:00
AM 5
09-01-2014
01:00 AM 2 Rejected Rejected
4
02-01-2014 12:00
AM 15
16-02-2014
12:00 AM 4 08-J an-2014 25-01-2014
5
03-01-2014 12:00
AM 12
19-02-2014
12:00 AM 4 16-J an-2014 07-02-2014
6
04-01-2014 12:00
AM 4
07-02-2014
12:00 AM 1 29-J an-2014 01-02-2014
7
05-01-2014 12:00
AM 10
18-01-2014
8 05-01-2014 12:00 3 17-02-2014 2 02-Feb- 05-02-2014


AM 12:00 AM 2014
9
07-01-2014 12:00
AM 10
08-03-2014
12:00 AM 2
06-Feb-
2014 16-02-2014
10
08-01-2014 12:00
AM 7
16-03-2014
12:00 AM 2
08-Feb-
2014 16-02-2014

Table 2: mEDF algorithmwith same requests as in Table-1
Lid Req. Arrival
Duration
in days Deadline
Num
Nodes
Response
time Completion time
1
01-01-2014 12:00
AM 10
14-01-2014
2
01-01-2014 12:00
AM 8
12-01-2014
12:00 AM 2 01-01-2014 08-01-2014
3
01-01-2014 12:00
PM 5
09-01-2014
12:00 PM 2 01-01-2014 06-J an-14
4
02-01-2014 12:00
AM 15
16-02-2014
12:00 AM 4 10-01-2014 26-J an-14
5
03-01-2014 12:00
AM 12
19-02-2014
12:00 AM 4 16-01-2014 10-Feb-14
6
04-01-2014 12:00
AM 4
07-02-2014
12:00 AM 1 07-01-2014 11-J an-14
7
05-01-2014 12:00
AM 10
18-01-2014
8
05-01-2014 12:00
AM 3
17-02-2014
12:00 AM 2 27-01-2014 30-01-2014
9
07-01-2014 12:00
AM 10
08-03-2014
12:00 AM 2 01-02-2014 10-02-2014
10
08-01-2014 12:00
AM 7
16-03-2014
12:00 AM 2 26-01-2014 02-02-2014

Figure 2 Graph showing lease id and response times of DLS and mEDF algorithms.

Figure 3 Graph showing lease id and completion times of DLS and mEDF algorithms.


6. CONCLUSION AND FUTURE WORK
mEDF algorithm implementation shows improvement in response time and completion time of various leases. This
algorithmcan be very beneficial to cloud hosts and consumers. Proposed algorithmwill automatically reject the leases,
which cannot be completed in time thus it saves resources for other leases and improves response time and completion time
of lease requests of consumers. Leases with scientific workloads will get most benefitted by this algorithm.
Future work may use experimenting same algorithmwith backfilling algorithms conservative and off modes and provide a
comparative study of all deadline based algorithms. This work also suggests taking CRI score [7] along with mEDF
algorithmto improve resource utilization and revenue at cloud host.

References
[1] The Oxford Dictionary Website. [Online]. Available: http://www.oxforddictionaries.com/definition/english/cloud-
computing.
[2] S. Abrishami, and M. Naghibzadeh. Deadline-constrained workflow scheduling in software as a service cloud,
Scientia Iranica XIX (3), 680-689, 2012.
[3] Arshad, Junaid, Paul Townend, and Jie Xu. "Quantification of security for compute intensive workloads in clouds,"
In Proceedings of the 15th International Conference on Parallel and Distributed Systems (ICPADS), 2009, pp. 479-
486., 2009.
[4] V. Shrivastava, and D. S. Bhilare. CBUD Micro: A Micro Benchmark for Performance Measurement and Resource
Management in IaaS Clouds,International Journal of Emerging Technolgy and Advanced Engineering, III(11), pp.
433-437, 2013.
[5] Ostermann, Simon, Alexandria Iosup, Nezih Yigitbasi, Radu Prodan, Thomas Fahringer, and Dick Epema. "A
performance analysis of EC2 cloud computing services for scientific computing." In Cloud Computing, pp. 115-131.
Springer Berlin Heidelberg, 2010.
[6] Sotomayor, Borja, Rubn S. Montero, Ignacio M. Llorente, and Ian Foster. "Virtual infrastructure management in
private and hybrid clouds." Internet Computing, IEEE XIII (5), pp. 14-22, 2009.
[7] V. Shrivastava and D. S. Bhilare, CRI: A Novel Rating Based Leasing Policy and Algorithm for Efficient Resource
Management in IaaS Clouds,International J ournal of Computer Science and Information Technologies, V (3), pp.
4226-4230, 2014.
[8] H. Mehta, P. Kanungo, and M. Chandwani."Performance enhancement of scheduling algorithms in web server
clusters using improved dynamic load balancing policies, In Proceedings of the 2nd National Conference,
INDIACom-2008 Computing For Nation Development, pp. 651-656, 2008.
[9] Q. Zhang, L. Cheng, and R. Boutaba. Cloud computing: state-of-the-art and research challenges, Journal of
internet services and applications I (1), pp. 7-18, 2010.
[10] C. Vecchiola, C. Rodrigo, N. Calheiros, D. Karunamoorthy, and R. Buyya. Deadline-driven provisioning of
resources for scientific applications in hybrid clouds with Aneka, Future Generation Computer Systems XXVIII (1)
pp. 58-65, 2012.
[11] S. Pandey, L. Wu, M. G. Siddeswara, and R.Buyya. A particle swarm optimization-based heuristic for scheduling
workflow applications in cloud computing environments, In Proceedings of the 24th IEEE International Conference
on Advanced Information Networking and Applications (AINA), pp. 400-407, 2010.
[12] The OpenNebula Website. [Online]. Available: http://opennebula.org/.
[13] Eucalyptus Website. [Online]. Available: https://www.eucalyptus.com/eucalyptus-cloud/iaas [Accessed: Jan. 2, 2014].
[14] OpenStack Website. [Online]. Available: https://www.openstack.org/ [Accessed: Jan. 1, 2014].
[15] The Haizea Website. [Online]. Available: http://haizea.cs.uchicago.edu/ [Accessed: Jan. 1, 2014]..
[16] A. Nathani, S. Chaudhary, and G. Somani. Policy based resource allocation in IaaS cloud, Future Generation
Computer Systems XXVIII (1), 94-103, 2012.
[17] B. Sotomayor. The Haizea Manual on Haizea, [Online]. Available:
http://haizea.cs.uchicago.edu/haizea_manual.pdf [Accessed: Jan. 1, 2014]..
[18] J. Akhani, S. Chuadhary, and G. Somani. Negotiation for resource allocation in IaaS cloud, In Proceedings of the
Fourth Annual ACM Bangalore Conference, pp. 15. ACM, 2011.
[19] V. Shrivastava, and D. S. Bhilare. Algorithms to Improve Resource Utilization and Request Acceptance Rate in
IaaS Cloud Scheduling, International Journal of Advanced Networking & Applications, III (5), pp. 1367-1374,
2012.
[20] B. Sotomayor, K. Keahey, and I. Foster. Combining batch execution and leasing using virtual machines, In
Proceedings of the 17th international symposium on High performance distributed computing, pp. 87-96, 2008.
[21] B. Sotomayor, R. S. Montero, I.M. Llorente, and I. Foster. Capacity leasing in cloud systems using the opennebula
engine," In Workshop on Cloud Computing and its Applications, III. 2008.


[22] B. Sotomayor, R.S. Montero, I. M. Llorente, and I. Foster. Resource leasing and the art of suspending virtual
machine, In Proceedings of 11th IEEE International Conference on High Performance Computing and
Communications, HPCC'09, pp. 59-68. 2009.
[23] D. Vora, S. Chaudhary, M. Bhise, V. Kumar, and G. Somani. Allocation of slotted deadline sensitive leases in
infrastructure cloud,, Distributed Computing and Internet Technology, pp. 242-252. Springer Berlin Heidelberg,
2012.
[24] M. Mao and M. Humphrey. Auto-scaling to minimize cost and meet application deadlines in cloud workflows, In
Proceedings of 2011 International Conference for High Performance Computing, Networking, Storage and Analysis,
pp. 49, 2011.
[25] J. Yu, R. Buyya, and C. K. Tham. Cost-based scheduling of scientific workflow applications on utility grids, In
Proceedings of the First International Conference on e-Science and Grid Computing, 2005., pp. 8, 2005.
[26] A. Verma, and S. Kaushal. Deadline and budget distribution based cost-time optimization workflow scheduling
algorithm for cloud, In Proceedings of the IJCA on International Conference on Recent Advances and Future
Trends in Information Technology (iRAFIT12), pp. 1-,. 2012.
[27] P. Brucker, Scheduling algorithms, 5th ed, Berlin: Springer-Verlag, 2007.
[28] The Wikipedia Website. [Online]. Available:http://en.wikipedia.org/wiki/Scheduling_algorithm, [Accessed: Jan. 1,
2014].

AUTHORS

Vivek Shrivastava is pursuing Ph. D. in CS from DAVV under the guidance of Dr. D.S. Bhilare. He is
M.Tech. (Computer Sc.), MCA, and UGC-NET qualified in Computer Sc. & Applications. Working as an
Assistant Professor in International Institute of Professional Studies, Devi Ahilya University, Indore. His areas
of interest are Cloud Computing, Ubiquitous Computing and Information Security.

D.S. Bhilare received his Ph.D. (Computer Science), M.Tech. (Computer Sc.), M.Phil. (Computer Sc.) and
MBA from Devi Ahilya University, Indore. Worked as a senior project leader for ten years in the industry and
developed various business applications for different industries. Since last twenty years, working in the
University as a Senior Manager & Head Computer Centre. He has presented several papers in national and international
conferences and published in reputed journals and international conference proceedings in the area of Information
Security.

Medf: Deadline Driven Algorithm For Minimizing Response Time and Completion Time in Iaas Clouds

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Medf: Deadline Driven Algorithm For Minimizing Response Time and Completion Time in Iaas Clouds

Transféré par

Droits d'auteur :

Formats disponibles

International Journal of Application or Innovation in Engineering& Management (IJAIEM)

Web Site: www.ijaiem.org Email: editor@ijaiem.org

Vous aimerez peut-être aussi