Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi,...

18
upporting Load Balancing for Distribute Data-Intensive Applications Leonid Glimcher, Vignesh Ravi, and Gagan Agrawal Department of ComputerScience and Engg. The Ohio State University Columbus, Ohio - 43210

Transcript of Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi,...

Page 1: Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi, and Gagan Agrawal Department of ComputerScience and.

Supporting Load Balancing for DistributedData-Intensive Applications

Leonid Glimcher, Vignesh Ravi, and Gagan AgrawalDepartment of ComputerScience and Engg.

The Ohio State UniversityColumbus, Ohio - 43210

Page 2: Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi, and Gagan Agrawal Department of ComputerScience and.

Outline

• Introduction• Motivation• FREERIDE-G Processing Structure• Run-time Load Balancing System• Experimental Results• Conclusions

April 21, 2023 2

Page 3: Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi, and Gagan Agrawal Department of ComputerScience and.

Introduction

• Growing abundance of data– Sensors, scientific simulations and business

transactions

• Data Analysis– Translate raw data into knowledge

• Grid/Cloud Computing– Enables distributed processing

April 21, 2023 3

Page 4: Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi, and Gagan Agrawal Department of ComputerScience and.

Motivation

• Resources are geographically distributed– Data nodes

– Compute nodes

– Middleware user

• Remote data analysis is important

• Heterogeneity of resources– Difference in network bandwidth

– Difference in compute power

April 21, 2023 4

Data Nodes

Compute Nodes

Middleware user

Grid/Cloud Environment

Page 5: Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi, and Gagan Agrawal Department of ComputerScience and.

FREERIDE-G Processing Structure(Framework for Rapid Implementation of Datamining Engines –

Grid)

April 21, 2023 5

While( ) {

forall( data instances d) {

(I , d’) = process(d)

R(I) = R(I) op d’

}

…….

}

• A Map-reduce like system

• Remote data analysis

Middleware API

• Process

• Reduce

• Global CombineReduction Object

Page 6: Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi, and Gagan Agrawal Department of ComputerScience and.

A Real-time Grid/Cloud Scenario

April 21, 2023 6

A B

C DCompute

Data

Page 7: Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi, and Gagan Agrawal Department of ComputerScience and.

Run-time Load Balancing

April 21, 2023 7

Two factors of load imbalance• Computational factor, w1• Remote data transfer (wait time), w2Case 1: w1 > w2Case 2: w2 > w1We use sum of weights to account for both

thecomponents

Page 8: Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi, and Gagan Agrawal Department of ComputerScience and.

Dynamic Load Balancing Algorithm

April 21, 2023 8

Consider every chunk, Ci

CalculateCompute cost, Cc

CalculateData transfer cost, Tc

Input Bandwidth matrix, W1 & W2

Total cost = W1*Cc + W2*Tc

If Total cost < Min

Update Min Assign Ci to Pj

Page 9: Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi, and Gagan Agrawal Department of ComputerScience and.

Experimental Setup

Settings

• Organizational Grid

• Wide Area Network (WAN)

Goals are to evaluate

• Scalability

• Dynamic Load balancing overhead

• Adaptability to scenarios

– compute bound,

– I/O bound,

– WAN setting

Applications

• K-means

• Vortex Detection

April 21, 2023 9

Page 10: Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi, and Gagan Agrawal Department of ComputerScience and.

April 21, 2023 10

Scalability and Overhead of Dynamic Balancing

Vortex detection

• 14.8 GB data• Organizational

setting• Bandwidth

– 50mb/sec– 100mb/sec

• 31% benefit• Overhead within

10%

April 21, 2023 10

Page 11: Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi, and Gagan Agrawal Department of ComputerScience and.

Model Adaptability – Compute Bound Scenario

Kmeans clustering• 25.6 GB data• Bandwidth

– 50 MB – 200 MB

Best result• 75-25 combination• skewed towards work load

componentInitial (unbalanced) overhead• 57% over balancedDynamic overhead• 5% over balanced

April 21, 2023 11

Ideal CaseDynamic case

ComputeData transfer

Page 12: Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi, and Gagan Agrawal Department of ComputerScience and.

Model Adaptability – I/O Bound Scenario

April 21, 2023 12

Kmeans clustering• 25.6 GB data• Bandwidth

– 15 mb/s– 60 mb/s

Best result• 25-75 combination• skewed towards data

transfer componentInitial (unbalanced)

overhead• 40% over balancedDynamic overhead• 4% over balanced

Page 13: Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi, and Gagan Agrawal Department of ComputerScience and.

Model Adaptability – WAN setting

Vortex Detection • 14.6 GBBest result• 25-75 combination results

in lowest overhead (favoring data delivery component)

Unbalanced configuration• 20% overhead over

balancedOur approach• Overhead reduced to 8%

April 21, 2023 13

0500

1000

Execu

tio

n T

ime (

sec)

4 compute - 8 data 8 compute - 8 data

staric balancing 0-100 25-75

50-50 75-25 100-0

unbalanced

Page 14: Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi, and Gagan Agrawal Department of ComputerScience and.

Conclusions

• Dynamic load balancing solution for grid environments

• Both workload and data transfer factors are important

• Scalability is good and overheads are within 10%

• Adaptable to compute-bound, I/O bound, and WAN settings

April 21, 2023 14

Page 15: Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi, and Gagan Agrawal Department of ComputerScience and.

April 21, 2023 15

Thank You!

Questions?

Contacts:Leonid Glimcher [email protected]

Vignesh Ravi - [email protected]

Gagan Agrawal - [email protected]

Page 16: Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi, and Gagan Agrawal Department of ComputerScience and.

[email protected]. 16

DataGrid Lab

Setup 1: Organizational GridData hosted on Opteron 250’s

Processed on Opteron 254’s

2 clusters connected through two 10 GB optical fibers

Both clusters within same city (0.5 mile apart)

Evaluating:

Scalability

Adaptability

Integration overheadCompute cluster (cse-ri)

Repository cluster (bmi-ri)

Page 17: Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi, and Gagan Agrawal Department of ComputerScience and.

[email protected]. 17

DataGrid Lab

Setup 2: WANData Repository:

Opteron 250’s (OSU)

Opteron 258’s (Kent St)

Processed on Opteron 254’s

No dedicated link between processing and repository clusters

Evaluating:

Scalability

Adaptability

Compute cluster (OSU)

Repository cluster (Kent ST)

Repository cluster (OSU)

Page 18: Supporting Load Balancing for Distributed Data-Intensive Applications Leonid Glimcher, Vignesh Ravi, and Gagan Agrawal Department of ComputerScience and.

FREERIDE-G System Design

April 21, 2023 18