Upload
julia-diane-chambers
View
218
Download
0
Tags:
Embed Size (px)
Citation preview
A Prediction-based Fair Replication Algorithm in Structured P2P Systems
Xianshu Zhu, Dafang Zhang, Wenjia Li, Kun HuangXianshu Zhu, Dafang Zhang, Wenjia Li, Kun Huang
Presented by: Xianshu ZhuPresented by: Xianshu ZhuCollege of Computer & Communication, Hunan University, P.R.ChinaCollege of Computer & Communication, Hunan University, P.R.China
OutlineOutline
IntroductionIntroductionContributionContributionPFR (Prediction-based Fair PFR (Prediction-based Fair Replication)Replication)Performance EvaluationPerformance EvaluationConclusion and Future WorkConclusion and Future Work
IntroductionIntroduction
Query Hotspot Query Hotspot
Structured Peer-to-Peer NetworkStructured Peer-to-Peer Network
Summary of Replication SchemesSummary of Replication Schemes
Query HotspotQuery Hotspot
FF
GG
II
JJ
CCDDEE
HH
BB
FilFilee
FilFilee
Query Hotspot: the number of requests for Query Hotspot: the number of requests for popular objects increases dramatically, and popular objects increases dramatically, and leads to consequent dropping queries and leads to consequent dropping queries and severe performance failures.severe performance failures.
Query Hotspot
Structured P2P Structured P2P NetworkNetwork
AdvantageAdvantage :: - Scalability - Scalability - Efficient Searching- Efficient Searching
DisadvantageDisadvantage :: The Implementation of StructureThe Implementation of Structured P2P Network Assumes that All Data Items are of d P2P Network Assumes that All Data Items are of the Same Popularity. No Mechanism Can Handle Hthe Same Popularity. No Mechanism Can Handle Hotspot Problemotspot Problem
Replication SchemesReplication Schemes
Basic IdeaBasic Idea :: - Distribute Replicas of the Popular Data Items to Vari- Distribute Replicas of the Popular Data Items to Vari
ous Light-loaded Nodesous Light-loaded Nodes - - FairlyFairly Distribute Load onto Each Node.Distribute Load onto Each Node.
When Apply Replication Technique: When Apply Replication Technique: -- Replica Creation: Time, Number, LocationReplica Creation: Time, Number, Location -- Replica UtilizationReplica Utilization
Replication SchemesReplication Schemes
Classification According to Replica Location:Classification According to Replica Location:
- Path Replication- Path Replication
- Owner Replication- Owner Replication
- Random Replication- Random Replication
AA BB CC DD EE FFFilFilee
FilFilee
FilFilee
FilFilee
FilFilee
FilFilee
FilFilee
FilFilee
FilFilee
FilFilee
FilFilee
FilFilee
High Replication High Replication OverheadOverhead
Replication SchemesReplication Schemes
AA BB CC DD EE FFFile File AA
File File AA
1.New Query Hotspot1.New Query Hotspot2.Low Replication Speed2.Low Replication Speed
Classification According to Replica Location:Classification According to Replica Location: - Path Replication- Path Replication - Owner Replication: Gopalakrishnan proposed LAR- Owner Replication: Gopalakrishnan proposed LAR - Random Replication- Random Replication
File BFile BFile BFile B File DFile DFile DFile D File DFile DFile DFile DFile BFile BFile BFile BFile AFile AFile AFile A
Replication SchemesReplication Schemes
AA BB CC DD EE FFFilFilee
FilFilee
FilFilee
FilFilee
FilFilee
FilFilee
FilFilee
FilFilee
Classification According to Replica Location:Classification According to Replica Location:
- Path Replication- Path Replication
- Owner Replication- Owner Replication
- Random Replication- Random Replication
OutlineOutline
IntroductionIntroductionContributionContributionPFR (Prediction-based Fair PFR (Prediction-based Fair Replication)Replication)Performance EvaluationPerformance EvaluationConclusion and Future WorkConclusion and Future Work
Contribution
Design Goals:Design Goals: - Dropped Queries by Only Introducing - Dropped Queries by Only Introducing
Minimum Replication OverheadMinimum Replication Overhead - - Minimize the Drawbacks of LAR AlgorithmMinimize the Drawbacks of LAR Algorithm
(Owner Replication)(Owner Replication)
Prediction-based Fair Replication Algorithm Prediction-based Fair Replication Algorithm (PFR) that Can Almost Fairly Distribute Load (PFR) that Can Almost Fairly Distribute Load onto Each Node, So As to Meet the Above onto Each Node, So As to Meet the Above Design Goal. Design Goal.
Contribution
Fairness Goal of PFR -Adaptively Determine the Replication Speed and Replication
Location According to Node’s Predicted Load Fraction
AA BB CC DD EE FF GG
OutlineOutline
IntroductionIntroductionContributionContributionPFR (Prediction-based Fair PFR (Prediction-based Fair Replication)Replication)Performance EvaluationPerformance EvaluationConclusion and Future WorkConclusion and Future Work
Predict(n+1)Predict(n+1)
PFR- PFR- Appropriate Replication Appropriate Replication
TimeTime
To keep the System Performance at a High Level, PrTo keep the System Performance at a High Level, Preventive Actions Should be Taken Before Query Hoteventive Actions Should be Taken Before Query Hotspot Really Happensspot Really HappensPeriod Exponential Weight Prediction AlgorithmPeriod Exponential Weight Prediction Algorithm
Predict(n+1)=Current(n) + PredictDiff(n+1)Predict(n+1)=Current(n) + PredictDiff(n+1)
12nn+1
n-1
Current Current TimeTime
Predicted Possible Traffic Difference Between nth and (n+1)th Predicted Possible Traffic Difference Between nth and (n+1)th IntervalInterval
Period Exponential Weight Prediction Period Exponential Weight Prediction AlgorithmAlgorithm
- Only Incurs Low Computation Overhead- Only Incurs Low Computation Overhead - Applicable to Online Prediction- Applicable to Online Prediction
Our Replication Strategy is Set Based on Our Replication Strategy is Set Based on The Predicted loadThe Predicted load
PFR- Appropriate Replication PFR- Appropriate Replication
TimeTime
Replication Speed:Replication Speed:
AA BB CC DD EE FFFilFilee
FilFilee
FilFilee
FilFilee
FilFilee
FilFilee
FilFilee
FilFilee
3/63/6
Replication SpeedReplication Speed=(the Number of Nodes Chosen =(the Number of Nodes Chosen to Hold Replicas)/(the Number of All Nodes that to Hold Replicas)/(the Number of All Nodes that Have Encountered Along the Query Path)Have Encountered Along the Query Path)
PFR- Fairly-decided Replication PFR- Fairly-decided Replication SpeedSpeed
Replication Level:Replication Level:
NN
N/2N/23N/43N/4
N/4N/411DON’T create DON’T create replicasreplicas
N: Total Number of Nodes Along a Query N: Total Number of Nodes Along a Query PathPath
PFR- Fairly-decided Replication PFR- Fairly-decided Replication SpeedSpeed
Replication Replication SpeedSpeed
Predicted Load Predicted Load FractionFraction
(0.5)(0.5)
(0.3)(0.3)
(0.6(0.6))
(0.7(0.7))
(0.8(0.8))(1)(1)
Node Homogeneity
PFR- Replication & Replica PFR- Replication & Replica UtilizationUtilization
AA BB CC DD EE FF
GGC: C: FileFile
C: C: FileFile
F:0.25F:0.25E:0.15E:0.15F:0.25F:0.25E:0.15E:0.15
F:0.25F:0.25
D:0.3D:0.3C:0.55C:0.55
E:0.15E:0.15F:0.25F:0.25
D:0.3D:0.3B:0.3B:0.3C:0.55C:0.55
E:0.15E:0.15F:0.25F:0.25
D:0.3D:0.3
A:0.9A:0.9B:0.3B:0.3C:0.55C:0.55
E:0.15E:0.15F:0.25F:0.25
D:0.3D:0.3
RS:N/4=1RS:N/4=1
A: FileA: FileA: FileA: File A: FileA: FileA: FileA: File
A: FileA: FileA: FileA: File
A: FileA: FileA: FileA: File A: FileA: FileA: FileA: File
RS:NRS:N
E:CE:CE:CE:C
E:CE:C
B,D,E,F:AB,D,E,F:A
B,D,E,F:AB,D,E,F:AB,D,E,F:AB,D,E,F:A
B,D,E,F:AB,D,E,F:A
B,D,E,F:AB,D,E,F:A
B,D,E,F:AB,D,E,F:A
D:AD:A N=6N=6
OutlineOutline
IntroductionIntroductionContributionContributionPFR (Prediction-based Fair PFR (Prediction-based Fair Replication)Replication)Performance EvaluationPerformance EvaluationConclusion and Future WorkConclusion and Future Work
Performance EvaluationPerformance Evaluation
Highly modified Chord Simulator from MIT and Highly modified Chord Simulator from MIT and LAR Implementation CodeLAR Implementation Code ::
System SizeSystem Size 10001000 The Time Each The Time Each Network hop Network hop takestakes
25ms25ms
Number of Number of datadata
3276732767 Average system Average system loadload
25%25%
Node capacityNode capacity 10 per 10 per secsec
Number of Number of Queries Queries Generate per Generate per SecSec
500500
Node’s queue Node’s queue lengthlength
3232 Prediction Prediction intervalinterval
1s1s
Performance EvaluationPerformance Evaluation
Number of Queries Dropped Over Number of Queries Dropped Over TimeTime
28%28%
90% of the input queries are directed to 190% of the input queries are directed to 1 itemitem
LARLAR
PFRPFR
Performance EvaluationPerformance Evaluation
Total Number of Documents ReplicatedTotal Number of Documents Replicated
LARLAR
PFRPFR
Performance EvaluationPerformance Evaluation
Total Number of Finger Tables Total Number of Finger Tables ReplicatedReplicated
LARLAR
PFRPFR
Performance EvaluationPerformance Evaluation
Total Number of Replica Location Hints Total Number of Replica Location Hints CreatedCreated
PFRPFR
LARLAR
OutlineOutline
IntroductionIntroductionContributionContributionPFR (Prediction-based Fair PFR (Prediction-based Fair Replication)Replication)Performance EvaluationPerformance EvaluationConclusion and Future WorkConclusion and Future Work
ConclusionConclusion
Prediction-based Fair Replication Algorithm Prediction-based Fair Replication Algorithm Can Conduct Fair Replication through:Can Conduct Fair Replication through:
- Appropriate Replication Time- Appropriate Replication Time - Fairly-decided Replication Speed- Fairly-decided Replication Speed - Fairly-decided Replication Location- Fairly-decided Replication Location - High Replica Utilization Rate- High Replica Utilization Rate
Performance Evaluation:Performance Evaluation: - Notably Decrease the Number of Dropped - Notably Decrease the Number of Dropped
QueriesQueries - Low Replication Overhead- Low Replication Overhead
Future WorkFuture Work
Taking Node Heterogeneity into ConsiderationTaking Node Heterogeneity into Consideration
Thank you!Thank you!