efficient scheduler for electronic funds tranfer (eft) scenarios

16
LoadEFT : Efficient Scheduler Proposal for Electronic Funds Transfer Companies Tiago Nascimento, Cristiano André da Costa, Rodrigo da Rosa Righi CLEI 2014 - Conferencia Latinoamericana em Informática Contact: [email protected]

Upload: rrrighi

Post on 15-Aug-2015

16 views

Category:

Economy & Finance


0 download

TRANSCRIPT

LoadEFT: Efficient Scheduler Proposal for Electronic Funds Transfer CompaniesTiago Nascimento, Cristiano André da Costa, Rodrigo da Rosa Righi

CLEI 2014 - Conferencia Latinoamericana em Informática

Contact: [email protected]

Agenda• Introduction• LoadEFT• Evaluation Methodology• Results• Conclusion

IntroductionElectronic medias for payment operations have become mainstream, instead of using money in currency paper and check.

After arriving in the service provider company, transactions are received by a switch that acts as a scheduler responsible for assigning them to processing machines.

IntroductionThe standard mechanism for mapping transactions to PMs is based on the Round-Robin algorithm, which schedules a list of resources in a circular fashion.

Round-Robin achieves the Optimum schedule when all transactions present a single type, with the same CPU and I/O requirements among themselves, and PMs are built with the same pieces of hardware and present the same software configuration.

T1 SchedulerT2T3T4T5

PM1

PM2

PM3

PM4

T1

T2

T3

T4

T5Homogeneous Transactions (T)

HomogeneousProcessing

Machines (PM)

Round-RobinOptimal

Scheduling

500 flops

1 GHz

T1 T5. . .

PM1 PM4. . .

Legend

Introduction

Nevertheless, the Round-Robin scheduling approach may not be the best alternative when either dynamic or heterogeneous systems are involved.

In spite of the disadvantages explained earlier, many transaction processing companies still use a Round-Robin-based scheduling.

T1 SchedulerT2T3T4T5

PM1

PM2

PM3

PM4

T1

T2

T3

T4

T5

Heterogeneous Transactions (T)

HeterogeneousProcessing

Machines (PM)GetLB

Scheduling

T1 T4

T2 T3 T5

PM1 PM3

PM2 PM4

500 flops

1 GHz

1000 flops

500 MHz

Legend

T1 SchedulerT2T3T4T5

PM1

PM2

PM3

PM4

T1

T2

T3

T4

T5Heterogenenous Transactions (T)

HomogeneousProcessing

Machines (PM)Round-RobinScheduling

1 GHzPM1 PM4. . .

T1 T4

T2 T3 T5 500 flops

1000 flops

Legend

Introduction

EFT works with heterogeneous transactions

Heterogeneity on transactions

Heterogeneity on resources

Introduction

T1Slice1

T1Slice2

T1Slice3

T2Slice2

T2Slice1

T1Slice1

T1Slice2

T1Slice3

T2Slice1

T2Slice2

Transactions Flow Dispatcher

ProcessingMachines

Example ofScheduling

T1Company1

T2Company1

T3Company1

T4Company2

T1Company1

T2Company1

T3Company1

T4Company2

Dispatcher

ProcessingMachines

Example ofScheduling

Transactions Flow

Technical Problems

Allocation for companies

Packet-based scheduling

Introduction

LoadEFT: Load balancing model for Electronic Funds Transfer Systems

Dynamic Data about transactions and processing machines

Performance center to save data about the targets

Simple prediction using moving averages

Overcome technical problems

LoadEFT

Internet

User Cluster Frontend server

Virtual Servers192.168.0.1: 80 Cluster Web192.168.0.1: 443 Cluster SSL192.168.0.1: 23 Cluster Telnet

Machine172.16.1.11

Services172.16.1.11:80

Machine172.16.1.12

Machine172.16.1.13

Machine172.16.1.14

Services172.16.1.12:80

172.16.1.12:443

Services172.16.1.13:80172.16.1.13:23

Services172.16.1.14:80172.16.1.14:23

172.16.1.14:443

External Network: 192.168.0.10Internal Network: 172.16.1.1

Machine192.168.0.2

Idea: Use of Virtual Servers - Single point of entrance

LoadEFT

Driver Load Balancer

Driver Authorizer

Driver Authorizer

Driver Authorizer

Performance Center

Event Manager

Scalability Control

AuxiliaryServices

Transactions Waiting

ResponseAnswered

TransactionsAuthorizers

PerformanceShared-Memory

Region

MTD Scheduler

Load Balancing

Socket ConnectionAccess to the Shared-Memory

Communication FlowLegend

Architecture

Evaluation Methodology

4800 transactions

Cluster Type Host 1 Host 2 Host 3 Host 4

Homogeneous 20 20 20 20

Heterogeneous 10 20 40 10

Debit and credit transactions

Timeout of 7 seconds

Conditions

ResultsHomogeneous cluster

Round-Robin

LoadEFTMapping of transactions

ResultsHeterogeneous cluster

Round-Robin

LoadEFTMapping of transactions

Results

Number of both successfully answered transactions and timeouts

ConclusionImportance of load-balancing at runtime

Time is one of the most important factors on users’ experience

EFT is a heterogeneous system, so Round-Robin is not the most suitable option for such systems.

Scheduling Scheduling calculus

Scheduling Quality

LoadEFT presents better performance on

heterogeneous resources

= +

LoadEFT: Efficient Scheduler Proposal for Electronic Funds Transfer CompaniesTiago Nascimento, Cristiano André da Costa, Rodrigo da Rosa Righi

CLEI 2014 - Conferencia Latinoamericana em Informática

Contact: [email protected]