Green-CM: Energy efficient contention management for Transactional - - PowerPoint PPT Presentation

Green-CM: Energy efficient contention management for Transactional Memory

Shady Alaa Paolo Romano – INESC-ID/IST Mats Brorsson - KTH

Agenda

• Introduction
• Related work
• Architecture
• Green-CM
• Evaluation
• Conclusion

ICPP 2015 - Green-CM 2

Introduction

• Multicores are everywhere

– Complex programming

• Locks
• Deadlocks

– Transactional memory

• Atomics blocks
• Transparent from programmer

Main memory

Core 1 Core 2 Core 3 Core 4

atomic{ if(bal>amount) withdraw(amount); }

ICPP 2015 - Green-CM 3

Introduction

• Energy efficiency

– First order design choice – Battery based devices – Data centers

• Goal

– Energy efficient transactional memory in terms of both energy and performance

ICPP 2015 - Green-CM 4

Introduction

• Contention Manager

– minimize contention – which transaction to abort – when to restart an aborted transaction

• Energy efficiency:

– wait implementation – DVFS

ICPP 2015 - Green-CM 5

Related work

• Few work in literature

– Mainly HTM

• Clock gating processors upon abort

– Lowering frequency upon abort

• Using simulator
• Studies

– HTM consume lower energy

• Does not fit all workloads

– Need for adaptability

• Using DVFS in TM

– Fastlane

• Designed for low number of threads

ICPP 2015 - Green-CM 6

Architecture

Asymmetric* Conten.on Manager* Tx*abort* (no.*of*retries,* core*on*which* tx*is*execu.ng)* Throughput*

*

Energy* Controller* Hybrid* Wait** Implementa.on* backEoff* dura.on* Tuning*of* Β* Tuning*of* α,*Τ* * End** backEoff* Restart* Tx*

ICPP 2015 - Green-CM 7

Architecture

Asymmetric* Conten.on Manager* Tx*abort* (no.*of*retries,* core*on*which* tx*is*execu.ng)* Throughput*

*

Energy* Controller* Hybrid* Wait** Implementa.on* backEoff* dura.on* Tuning*of* Β* Tuning*of* α,*Τ* * End** backEoff* Restart* Tx*

ICPP 2015 - Green-CM 8

Implementing waits

• Building block for contention managers
• Drastic effect on energy consumption
• Can be implemented in two ways:

– Busy waiting – sleeping

ICPP 2015 - Green-CM 9

Implementing waits

• Busy waiting

– Fine granularity – Similar to real actual work

• Sleeping

– Coarse granularity – Low energy consumption – expensive

ICPP 2015 - Green-CM 10

Implementing waits

• Hybrid approach

– Either busy wait or sleep

• Adaptive fashion

– How to determine the threshold

• Cost of sleep

ICPP 2015 - Green-CM 11

Implementing waits

1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 100 1000 10000 100000 1x106 1x107 EDP / best EDP

• Static Thresholds

Intruder Kmeans Threshold

ICPP 2015 - Green-CM 12

No one size fits all

Architecture

Asymmetric* Conten.on Manager* Tx*abort* (no.*of*retries,* core*on*which* tx*is*execu.ng)* Throughput*

*

Energy* Controller* Hybrid* Wait** Implementa.on* backEoff* dura.on* Tuning*of* Β* Tuning*of* α,*Τ* * End** backEoff* Restart* Tx*

ICPP 2015 - Green-CM 13

Asymmetric CM

• DVFS

– Variable operating frequency

• Exploiting DVFS

– Boosting active threads – Reducing freq. of backing off threads

• Enabling DVFS

– Manual control is expensive – How to favor automatic boosting

P0 3.0 GHz P1 2.4 GHz P2 2.2 GHz P3 2.0 GHz P4 1.8 GHz P5 1.6 GHz P6 1.4 GHz

ICPP 2015 - Green-CM 14

Asymmetric CM

Linear backoff Exp. Backoff Exp. Backoff Exp. Backoff Linear backoff Exp. Backoff Exp. Backoff Exp. Backoff

• Linear backoff cores:

– Shorter backoff periods – Mainly busy waiting backoffs

• Exp. Backoff cores:

– Longer backoff periods – Mainly sleep waiting

• Favor boosting

– When enough cores are in sleep states

8 core processor

Busy wait Sleep Sleep Sleep Busy waiting Sleep Sleep Sleep

Boosted Sleep Sleep Sleep Boosted Sleep Sleep Sleep

ICPP 2015 - Green-CM 15

Asymmetric CM

• Increased contention?

– Cores not backing off exponentially

• Control number of cores to be boosted

ICPP 2015 - Green-CM 16

Asymmetric CM

0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 4 8 16 EDP / best EDP

• No. of Boosted Threads

Static No. of Boosted Threads Intruder Kmeans Genome Memcached STM7

ICPP 2015 - Green-CM 17

Architecture

Asymmetric* Conten.on Manager* Tx*abort* (no.*of*retries,* core*on*which* tx*is*execu.ng)* Throughput*

*

Energy* Controller* Hybrid* Wait** Implementa.on* backEoff* dura.on* Tuning*of* Β* Tuning*of* α,*Τ* * End** backEoff* Restart* Tx*

ICPP 2015 - Green-CM 18

Controller

• Online, lightweight
• Hill climbing
• Challenges:

– Collection of energy – Multi dimensional

• Different exploration strategies

– Stabilization – Random jumps

ICPP 2015 - Green-CM 19

Controller

• Tuning α (threshold for hybrid)

0.5 1 1.5 2 2.5 I n t r u d e r K m e a n s M e m c a c h e d S T M 7 A v e r a g e EDP / best EDP Benchmark

no stab stab stab jmp 1 stab jmp 10

ICPP 2015 - Green-CM 20

Controller

• Tuning β (no. of boosted threads)

0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 I n t r u d e r K m e a n s M e m c a c h e d S T M 7 A v e r a g e EDP / best EDP Benchmark

no stab stab stab jmp 1 stab jmp 10

ICPP 2015 - Green-CM 21

Controller

• Merging the learners

0.5 1 1.5 2 2.5 Intruder Kmeans Memcached STM7 Average EDP / best EDP Benchmark

Coupling the Tuners

independent stab jmp 1 stab – stab stab jmp 1 – stab stab jmp 10 – stab bidim stab jmp 1

ICPP 2015 - Green-CM 22

Evaluation

0.2 0.4 0.6 0.8 1 1.2 4 8 16 32 48 64 EDP-GreenCM / EDP Threads Intruder

ICPP 2015 - Green-CM 23

Evaluation

0.2 0.4 0.6 0.8 1 1.2 4 8 16 32 48 64 EDP-GreenCM / EDP Threads STM7

ICPP 2015 - Green-CM 24

Evaluation

0.2 0.4 0.6 0.8 1 1.2 4 8 16 32 48 64 EDP-GreenCM / EDP Threads Memcached

ICPP 2015 - Green-CM 25

Evaluation

spin no-asym asym % of total cores Intruder, 64 threads p0 p1 p2 p3 p4 p5 p6

ICPP 2015 - Green-CM 26

Conclusion

• Implementation of waits has a significant

impact on energy efficiency

• Experimental results (obtained on real

system) contradict previously published

• nes based on simulation
• Exploiting DVFS enhances energy

efficiency

• Self-tuning is needed to adapt to different

workloads

ICPP 2015 - Green-CM 27

THANK YOU

ICPP 2015 - Green-CM 28

Evaluation

ICPP 2015 - Green-CM 29

0.2 0.4 0.6 0.8 1 1.2 4 8 16 32 48 64 Energy-GreenCM / Energy Threads Intruder

Evaluation

ICPP 2015 - Green-CM 30

0.2 0.4 0.6 0.8 1 1.2 4 8 16 32 48 64 Time-GreenCM / Time Threads Intruder