scheduling of reducing cache pollution in multicore department of embedded software, korea...

16

Click here to load reader

Upload: gabriel-nicholson

Post on 19-Jan-2018

214 views

Category:

Documents


0 download

DESCRIPTION

Problem of Existing scheduling way Regardless of task and reduce cache contention Cache miss, cache pollution in Multicore Division- I/O task, general purpose task for performance improvement

TRANSCRIPT

Page 1: Scheduling of Reducing Cache Pollution in Multicore Department of Embedded software, Korea university 2011610012 Juho Lim

Scheduling of Reducing Cache Pollution in Multicore

Department of Embedded software, Korea university

2011610012 Juho Lim

Page 2: Scheduling of Reducing Cache Pollution in Multicore Department of Embedded software, Korea university 2011610012 Juho Lim

Problem Multicore

Shared memory, shared cache

Contention

Need to efficiently scheduling method in multicore

Performance down +increase energy consumption

Page 3: Scheduling of Reducing Cache Pollution in Multicore Department of Embedded software, Korea university 2011610012 Juho Lim

Problem of Existing scheduling way

Regardless of task and reduce cache contention

Cache miss, cache pollution in Multicore Division- I/O task, general purpose

task for performance improve-ment

Page 4: Scheduling of Reducing Cache Pollution in Multicore Department of Embedded software, Korea university 2011610012 Juho Lim

Animal Classification

• Turtle- simply do not make much use of shared last –level cache

• Sheep- task that low -cache access and cache miss rate

• Rabbits: low cache miss rate but frequently cache access tasks

• Devils: high cache miss rate and frequently cache access tasks.

How much effect to each core when scheduling with core pair which shared cache

Page 5: Scheduling of Reducing Cache Pollution in Multicore Department of Embedded software, Korea university 2011610012 Juho Lim

Animal Classification

Page 6: Scheduling of Reducing Cache Pollution in Multicore Department of Embedded software, Korea university 2011610012 Juho Lim

Resource-conscious scheduling

Task Activity Vec-tor

Task1: Frequently use resource

Task2: rarely use re-source

• Task Activity Vector divide Task1 and Task2 for simultane-ous run.

Page 7: Scheduling of Reducing Cache Pollution in Multicore Department of Embedded software, Korea university 2011610012 Juho Lim

Resource-conscious scheduling

Task1: Frequently use resource

Task2: rarely use re-source

Run Queu

e

Run Queu

e

Core

• Division two task type by degree how use resource • Create Two run queue• Insert two type task into run queue• According to Epoch, two run queue put the tasks to core by turns• This technique reduce resource contention so cores use fair re-

source

Page 8: Scheduling of Reducing Cache Pollution in Multicore Department of Embedded software, Korea university 2011610012 Juho Lim

Resource-conscious scheduling

Page 9: Scheduling of Reducing Cache Pollution in Multicore Department of Embedded software, Korea university 2011610012 Juho Lim

Devil Classification

Division – Devil area, Normal area

Devil- Disk I/O intensive tasks

Normal- other general tasks

IF Disk I/O > shared cache-> almost cache space became pollution state-> this make difficult to use of cache

So To find devil task is important!

Page 10: Scheduling of Reducing Cache Pollution in Multicore Department of Embedded software, Korea university 2011610012 Juho Lim

Devil Classification

Count I/O disk system call!-frequent? Devil-normal? Normal

So we classified scheduling task!

Division! Devil or Normal

Page 11: Scheduling of Reducing Cache Pollution in Multicore Department of Embedded software, Korea university 2011610012 Juho Lim

Devil Classification

Core2 Duo T9400 processor architecture

L2 cache state when run merge sort at core0

L2 cache state when run disk I/O at core1

L2 cache state when simul-taneous execute each core

Page 12: Scheduling of Reducing Cache Pollution in Multicore Department of Embedded software, Korea university 2011610012 Juho Lim

Test & Estimation

Cache miss rate(%

)

Page 13: Scheduling of Reducing Cache Pollution in Multicore Department of Embedded software, Korea university 2011610012 Juho Lim

Test & Estimation

Runtime (m

s)

Page 14: Scheduling of Reducing Cache Pollution in Multicore Department of Embedded software, Korea university 2011610012 Juho Lim

Conclusion

• Papers propose division tasks according to degree how use resource

• So Task division techniques are proposed Animal ClassificationResource-conscious schedulingDevil Classification

• For Effective Usage of cache, divide I/O intensive cache• And Schedule using divided task • We confirmed performance improvement

Page 15: Scheduling of Reducing Cache Pollution in Multicore Department of Embedded software, Korea university 2011610012 Juho Lim

Reference

[1] Deukhyon An, Junghan Kim, and Young Ik Eom, “Scheduling of Re-ducing Cache Pollution with I/O Tasks in Multicore Environment”, 2011.

[2] Sergey Zhuravlev, Sergey Blagodurov, Alexandra Fedorova, “Ad-dressing Shared Resource Contention in Multicore Processors via Scheduling, In ASPLOS, p.129-142, 2010.

[3] Yuejian Xie and Fabriel H.Loh, “Dynamic Classification of Program Memory Behaviors in CMPs”, In Proc. Of CMP-MSI, held in conjunction with ISCA-35,2008

[4] Andreas Merkel, Jan Stoess, Frank Bellosa, “Resource-conscious Scheduling for EnergyEfficiency on Multicore Processors”, EuroSys, p.153-166, 2010

Page 16: Scheduling of Reducing Cache Pollution in Multicore Department of Embedded software, Korea university 2011610012 Juho Lim