modeling and optimization of resource allocation in cloud · modeling and optimization of resource...
TRANSCRIPT
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Modeling and Optimization of Resource Allocation in CloudPhD Thesis Progress – Third Report
Atakan Aral
Thesis Advisor: Asst. Prof. Dr. Tolga Ovatman
Istanbul Technical University – Department of Computer Engineering
January 7, 2016
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Outline
1 IntroductionContribution to the ThesisTime Plan
2 Summary of the Previous Work
3 Literature Review
4 Cache Placement for Mobile Cloud ComputingProblem ModelingProposed Solution
5 Conclusion
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Contribution to the ThesisTime Plan
Outline
1 IntroductionContribution to the ThesisTime Plan
2 Summary of the Previous Work
3 Literature Review
4 Cache Placement for Mobile Cloud ComputingProblem ModelingProposed Solution
5 Conclusion
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Contribution to the ThesisTime Plan
Journal Submission
Submitted to Future Generation Computer Systems, ELSEVIER (IF: 2.786)SI: "Middleware Services for Heterogeneous Distributed Computing"First Decision Date: Nov 15, 2015 (Under review as of Jan 06, 2016)Also presented in IEEE 8th International Conference on Cloud Computing,CLOUD 2015
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Contribution to the ThesisTime Plan
Literature Review and Problem Modeling
Areas of interest:Mobile Cloud ComputingFog ComputingCloudlets, NanodatacentersSelf- and Context-aware Resource Management
Optimal Placement of Data Object Caches onto the CloudletsA distributed and context-aware algorithm
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Contribution to the ThesisTime Plan
Outline
1 IntroductionContribution to the ThesisTime Plan
2 Summary of the Previous Work
3 Literature Review
4 Cache Placement for Mobile Cloud ComputingProblem ModelingProposed Solution
5 Conclusion
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Contribution to the ThesisTime Plan
Gantt Chart
2015
7 8 9 10 11 12
TBM Evaluation
Manuscript Preparation
Journal Submission
Literature Review
Problem Modeling
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Outline
1 IntroductionContribution to the ThesisTime Plan
2 Summary of the Previous Work
3 Literature Review
4 Cache Placement for Mobile Cloud ComputingProblem ModelingProposed Solution
5 Conclusion
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Topology Based Mapping (TBM)
Main Idea
Map VM Clusters onto the federated cloud infrastructure based on their topology.
Decreases deployment latency (by placing VMs close to the broker)Decreases communication latency (by placing connected VMs to theneighbour data centers)Shortens execution time and increases throughputReduces resource costs (by balancing load and avoiding overload in any DC)
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
UML Activity Diagram
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Excluded Points
Geo-distributed user accessVirtual Machine or Data ReplicationUser mobilityVirtual Machine Migration
Topology Based Matching is a semi-centralized algorithmComplete utilization, capacity and topology information of the data centersand the network is available at all peers.
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Outline
1 IntroductionContribution to the ThesisTime Plan
2 Summary of the Previous Work
3 Literature Review
4 Cache Placement for Mobile Cloud ComputingProblem ModelingProposed Solution
5 Conclusion
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Mobile Cloud Computing
1 Computation is carried out in the cloud and the mobile device acts a thinclient.
Mobile elements are resource-poor relative to static elements.Mobile elements are more prone to loss, destruction, and subversion than staticelements.Mobile elements must operate under a much broader range of networkingconditions.
2 Nearby mobile devices form a cloud to assist each other in computationintensive tasks.
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Nano Data Centers
Small computation entities provided by ISPs on gateways/modems.Managed in a P2P architecture by the ISP.Main motivation is to reduce data center energy consumption.
Reuse already committed baseline powerAvoid cooling costsReduce network energy consumption
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Fog Computing
Main motivation is to leverage Internet of ThingsApplications that require very low latencyGeo-distributed applicationsFast mobile applications (vehicle, rail)Large-scale distributed control systems
Computation can be on high-end servers, edge routers, access points, set-topboxes, vehicles, sensors, mobile phonesCooperation between edge and core
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Cloudlets
"Data center in a box"Provided and owned by local businesses (e.g. coffee shops, offices)Allows code offloading using Virtual MachinesFall back to distant cloud or own resources of the mobile deviceLAN latency and bandwidthStores only cached data
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
Outline
1 IntroductionContribution to the ThesisTime Plan
2 Summary of the Previous Work
3 Literature Review
4 Cache Placement for Mobile Cloud ComputingProblem ModelingProposed Solution
5 Conclusion
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
Motivation
As the volume and velocity of the data in cloud is increasing, geographicaldistribution of where it is produced, processed and consumed is also gainingmore significanceMobile cloud computing offers a solution for the low-latency access tohigh-capacity computing resources.However, data is still mostly central and it is not feasible to replicate it in largenumber of geo-distributed locations.
Due to economical factorsDue to the limited storage capacity of the edge entitiesTo keep it consistent and available for analysis
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
Definition
Create caches of data objects on data centers and edge entitiesDecide the number and location of the caches based on:
Magnitude of user accessLocations of user accessCloud storage pricing
In an attempt to reduce:Data access latencyStorage cost
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
Issues and Requirements
Cost-Latency TradeoffCustomer preference for the level of aggression should be considered.
Complete topology information is no longer feasibleA distributed solution is necessary.
User access is dynamic and mobileThe solution must also be context-aware.
Edge entities have limited storage capacityConstraints must be respected.
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
Outline
1 IntroductionContribution to the ThesisTime Plan
2 Summary of the Previous Work
3 Literature Review
4 Cache Placement for Mobile Cloud ComputingProblem ModelingProposed Solution
5 Conclusion
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
Centralized Solutions
k-Medians Given a node set V with pairwise distance function d and servicedemands s(vj), ∀vj ∈ V , select up to k nodes to act as medians so asto minimize the service cost C(V , s, k).
C(V , s, k) =∑∀vj∈V
s(vj)d(vj ,m(vj))
Facility location Given a node set V with pairwise distance function d and servicedemands s(vj), ∀vj ∈ V and facility costs f (vj), ∀vj ∈ V , select a set ofnodes F to act as facilities so as to minimize the joint cost C(V , s, f )of acquiring the facilities and servicing the demand.
C(V , s, f ) =∑∀vj∈F
f (vj) +∑∀vj∈V
s(vj)d(vj ,m(vj))
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
Distributed Solution
Replication algorithm for the central storage:1 Create a cache for a data object in one of the neighbours.
Replication algorithm in the cache locations:1 Migrate the cache to one the neighbours.2 Duplicate the cache to one the neighbours.3 Remove the cache.
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
Sample Scenario
a
b
c
d
f
e
a1
a2
a3
a4
e1
b3
b2
b1
e4
e2
e3
f3
d1
f1
f2
d2
c1 c2
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
ITERATION 1d: User demand locations
a
b
c
d
f
e
a1
a2
a3
a4
e1
b3
b2
b1
e4
e2
e3
f3
d1
f1
f2
d2
c1 c2
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
ITERATION 1d: User demand received from c and f
a
b
c
d
f
e
a1
a2
a3
a4
e1
b3
b2
b1
e4
e2
e3
f3
d1
f1
f2
d2
c1 c2
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
ITERATION 1d: Cache creation decision
a
b
c
d
f
e
a1
a2
a3
a4
e1
b3
b2
b1
e4
e2
e3
f3
d1
f1
f2
d2
c1 c2
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
ITERATION 2f: Migration decision
a
b
c
d
f
e
a1
a2
a3
a4
e1
b3
b2
b1
e4
e2
e3
f3
d1
f1
f2
d2
c1 c2
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
ITERATION 2c: Duplication decision
a
b
c
d
f
e
a1
a2
a3
a4
e1
b3
b2
b1
e4
e2
e3
f3
d1
f1
f2
d2
c1 c2
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
ITERATION 3e: Migration decision
a
b
c
d
f
e
a1
a2
a3
a4
e1
b3
b2
b1
e4
e2
e3
f3
d1
f1
f2
d2
c1 c2
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
ITERATION 3a: Migration decision
a
b
c
d
f
e
a1
a2
a3
a4
e1
b3
b2
b1
e4
e2
e3
f3
d1
f1
f2
d2
c1 c2
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
ITERATION 3c: Removal decision
a
b
c
d
f
e
a1
a2
a3
a4
e1
b3
b2
b1
e4
e2
e3
f3
d1
f1
f2
d2
c1 c2
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
Inputs
Demand for each data object i from each neighbour j : Dij
Average latency for each data object i from each neighbour j : Lij
Latency from each node k to each neighbour j : Njk
Cost of storing each data object i at each neighbour and current location j : Cij
User provided level of aggression: A
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
Operation conditions
Create a cache of object i at neighbour j iff:
LijDijA > Cij
Remove the cache of the object i at k iff:∑∀j
(LijDijA) < Cik
Duplicate the cache of the object i from k to l iff:
LilDilA > Cil ∧∑∀j 6=l
(LijDijA) > Cik
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
Operation conditions
Migrate the cache of the object i from k to l iff:∑∀j
(LijDijA)−(∑
∀j 6=l
((Lij + Nkl)DijA
)+ (Lil − Nkl)DilA
)> Cil − Cik
A special case where ∃!j[Dij > 0]:
NklDilA > Cil − Cik
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
Possible Problems and Solutions
Multiple migrations/duplications are feasiblePrefer the option with the greatest benefit
Both migration and removal as feasiblePrefer migration
A costly node blocks the migration pathDynamic aggression level
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Problem ModelingProposed Solution
Contribution
There exists distributed VM replication methodsThe whole entity is replicated which is not feasible for big data.
There also exists distributed data storage methodsIn our model data is still stored centrally while caches are distributed.
As far as we are aware, all other studies apply a centralized approach.Not feasible in the case of mobile cloud computing where the topology is toolarge and dynamic.
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Outline
1 IntroductionContribution to the ThesisTime Plan
2 Summary of the Previous Work
3 Literature Review
4 Cache Placement for Mobile Cloud ComputingProblem ModelingProposed Solution
5 Conclusion
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Publications
Aral, A. and Ovatman, T. (2014). Improving resource utilization in cloudenvironments using application placement heuristics. In Proceedings of the4th International Conference on Cloud Computing and Services Science(CLOSER), pages 527–534.Aral, A. and Ovatman, T. (2015). Subgraph matching for resource allocation inthe federated cloud environment. In Proceedings of 8th IEEE InternationalConference on Cloud Computing (IEEE CLOUD), pages 1033–1036.Aral, A. and Ovatman, T. (2016). Network-Aware Embedding of VirtualMachine Clusters onto Federated Cloud Infrastructure. (Submitted to FGCSon 15-September-2015, under review as of 06-January-2016)
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Summary
Journal SubmissionLiterature ReviewProblem Modeling
Cache Placement for Mobile Cloud ComputingDistributed Context-Aware Algorithm
To reduce latencyTo decrease costs
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud
IntroductionSummary of the Previous Work
Literature ReviewCache Placement for Mobile Cloud Computing
Conclusion
Thank you for your time.
Atakan Aral Modeling and Optimization of Resource Allocation in Cloud