A Framework for Elastic Execution of Existing MPI Programs

1

A Framework for Elastic Execution of Existing MPI Programs

Aarthi Raveendran

Tekin Bicer Gagan Agrawal

A Framework for Elastic Execution of Existing MPI Programs 2

Motivation

Emergence of Cloud Computing• Including for HPC Applications

Key Advantages of Cloud Computing• Elasticity (dynamically acquire resources) • Pay-as-you model • Can be exploited to meet cost and/or time constraints

Existing HPC Applications• MPI-based, use fixed number of nodes

Need to make Existing MPI Applications Elastic

A Framework for Elastic Execution of Existing MPI Programs 3

Detailed Research Objective

To make MPI applications elastic • Exploit key advantage of Cloud Computing• Meet user defined time and/or cost constraints • Avoid new programming model or significant recoding

Design a framework for• Decision making • When to expand or contract

• Actual Support for Elasticity• Allocation, Data Redistribution, Restart

A Framework for Elastic Execution of Existing MPI Programs 4

Outline

Research ObjectiveFramework DesignRun time support modules Experimental Platform: Amazon Cloud ServicesApplications and Experimental EvaluationConclusion

A Framework for Elastic Execution of Existing MPI Programs 5

Framework components

A Framework for Elastic Execution of Existing MPI Programs 6

Framework design – Approach and Assumptions

Target – Iterative HPC Applications Assumption : Uniform work done at every

iterationMonitoring at the start of every few iterations of

the time-step loopCheckpointing and re- distribution Calculate required iteration time based on user

input

A Framework for Elastic Execution of Existing MPI Programs 7

Framework design - Modification to Source Code

Progress checked based on current average iteration time

Decision made to stop and restart if necessary Reallocation should not be done too frequentlyIf restarting is not necessary, the application

continues running

A Framework for Elastic Execution of Existing MPI Programs 8

Framework Design Execution flow

A Framework for Elastic Execution of Existing MPI Programs 9

Other Runtime Steps

Steps taken to perform scaling to a different number of nodes: Live variables and arrays need to be collected at the

master node and redistributed Read only need not be restored – just retrieve Application is restarted with each node reading the

local portions of the redistributed data.

A Framework for Elastic Execution of Existing MPI Programs 10

Runtime support modules Decision layer

Interaction with user and application program Constraints- Time or cost Monitoring the progress and making a decision Current work :

Measuring communication overhead and estimating scalability

Moving to large – type instances if necessary

A Framework for Elastic Execution of Existing MPI Programs 11

Framework design – Modification to Source Code

A Framework for Elastic Execution of Existing MPI Programs 12

Background – Amazon cloud

Services used in our framework : Amazon Elastic compute cloud (EC2)

Virtual images called instances Small instances : 1.7 GB of memory, 1 EC2 Compute Unit,

160 GB of local instance storage, 32-bit platform Large instances : 7.5 GB of memory, 4 EC2 Compute Units,

850 GB of local instance storage, 64-bit platform On demand , reserved , spot instances

A Framework for Elastic Execution of Existing MPI Programs 13

Background – Amazon cloud

Amazon Simple Storage Service (S3)

Provides key - value store Data stored in files Each file restricted to 5 GB Unlimited number of files

A Framework for Elastic Execution of Existing MPI Programs 14

Runtime support modulesResource allocator

Elastic execution Input taken from the decision layer on the number of

resources Allocating de- allocating resources in AWS

environment MPI configuration for these instances

Setting up of the MPI cluster Configuring for password less login among nodes

A Framework for Elastic Execution of Existing MPI Programs 15

Runtime support modules Check pointing and redistribution

Multiple design options feasible with the support available on AWSAmazon S3

Unmodified Arrays Quick access from EC2 instances Arrays stored in small sized chunks

Remote file copy Modified arrays (live arrays) File writes and reads

A Framework for Elastic Execution of Existing MPI Programs 16

Runtime support modules Check pointing and redistribution

Current design Knowledge of division of the original dataset

necessary Aggregation and redistribution done centrally on a

single node Future work

Source to source transformation tool Decentralized array distribution schemes

A Framework for Elastic Execution of Existing MPI Programs 17

Experiments

Framework and approach evaluated using Jacobi Conjugate Gradient (CG )

MPICH 2 used 4, 8 and 16 small instances used for processing

the data Observation made with and without scaling the

resources - Overheads 5-10% , which is negligible

A Framework for Elastic Execution of Existing MPI Programs 18

Experiments – Jacobi

A Framework for Elastic Execution of Existing MPI Programs 19

Experiments – Jacobi

A Framework for Elastic Execution of Existing MPI Programs 20

Experiments – Jacobi

Matrix updated at every iteration Updated matrix collected and redistributed at

node changeWorst case total redistribution overhead – less

than 2%Scalable application – performance increases with

number of nodes

A Framework for Elastic Execution of Existing MPI Programs 21

Experiments - CG

A Framework for Elastic Execution of Existing MPI Programs 22

Experiments - CG

A Framework for Elastic Execution of Existing MPI Programs 23

Experiments - CG

Single vector which needs to be redistributed Communication intensive application Not scalable Overheads are still low

A Framework for Elastic Execution of Existing MPI Programs 24

Conclusion

An overall approach to make MPI applications elastic and adaptable

An automated framework for deciding the number of instances for execution

Framework tested using 2 MPI applications showing low overheads during elastic execution