Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors Andreas Merkel, Jan Stoess, Frank Bellosa System Architecture Group KIT – The cooperation of Forschungszentrum Karlsruhe GmbH and Universität Karlsruhe (TH)
Memory Contention – a Problem on Multicores CPU Memory 2 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Memory Contention – a Problem on Multicores CPU Memory 3 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Memory Contention – a Problem on Multicores CPU CPU Memory 4 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Memory Contention – a Problem on Multicores CPU CPU CPU CPU Memory 5 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Memory Contention – a Problem on Multicores CPU CPU CPU CPU Memory CPU CPU CPU CPU 6 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Memory Contention – a Problem on Multicores CPU CPU CPU CPU CPU CPU CPU CPU Memory CPU CPU CPU CPU CPU CPU CPU CPU 7 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Memory Contention Intel Core2 Quad core1 core0 Bottleneck: memory bus idle stream core2 core3 idle idle Stall cycles, increased runtime core0 core1 stream stream 4.5 core2 core2 core3 idle idle idle 4 e core0 core1 m 3.5 stream stream i e t n c core2 core3 3 u n stream stream r 1 instance a } on 4 cores d t 2.5 s 2 instances e n z i 2 4 instances i l r a e m p 1.5 r o n 1 0.5 0 stream memory benchmark 8 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Impact of Resource Contention on Energy Efficiency Longer time to halt More static power Increasing importance of leakage 9 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Achieving Energy Efficiency by Scheduling Scheduler decides When Where In which combination At which frequency setting to execute tasks. What ist the most energy-efficient schedule? 10 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Achieving Energy Efficiency via Co-Scheduling Combination of tasks running together determines performance and energy efficiency Memory-bound + memory-bound: low energy efficiency Avoid memory bottleneck by combining memory- bound with compute bound tasks ➔ Co-schedule tasks with different characteristics energy avoid mem + comp efficiency contention 11 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Achieving Energy Efficiency via DVFS DVFS: Dynamic Voltage and Frequency Scaling Adapt processor frequency and voltage to task characteristics Memory-bound tasks: low frequency/voltage Compute-bound tasks: high frequency/voltage Multicore hardware limits options for frequency/voltage selection Often shared frequency/voltage domains ➔ Co-schedule similar tasks to select common best frequency and voltage energy use mem + mem efficiency DVFS 12 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Achieving Energy Efficiency use mem + mem DVFS energy efficiency avoid mem + comp contention 13 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Achieving Energy Efficiency use mem + mem DVFS energy efficiency avoid mem + comp contention 14 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Outline Analysis Resource contention Shared frequency/voltage domains Resource-conscious scheduling for energy efficiency OS task scheduling VM scheduling Frequency selection Evaluation Reduction of resource contention Increase in energy efficiency by 10 to 20% 15 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Analysis of Resource Contention on the Intel Core2 Quad Q6600 Contention for shared resources reduces energy efficiency Shared L2 caches (two cores) Shared memory interconnect (four cores) 16 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Resource Contention SPEC CPU 2006 4.5 4 3.5 e m i 1 instance e t 3 n c u 2 instances n r a 2.5 separate d t s e caches n z 2 i i 2 instances l r a e shared caches m p 1.5 r 4 instances o n 1 0.5 0 hmmer libquantum 17 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Resource Contention SPEC CPU 2006 4.5 4 3.5 e m i 1 instance e t 3 n c u 2 instances n r a 2.5 separate d t s e caches n z 2 i i 2 instances l r a e shared caches m p 1.5 r 4 instances o n 1 0.5 0 hmmer libquantum 18 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Resource Contention SPEC CPU 2006 4.5 4 3.5 e m i 1 instance e t 3 n c u 2 instances n r a 2.5 separate d t s e caches n z 2 i i 2 instances l r a e shared caches m p 1.5 r 4 instances o n 1 0.5 0 hmmer libquantum 19 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Resource Contention SPEC CPU 2006 4.5 4 3.5 e m i 1 instance e t 3 n c u 2 instances n r a 2.5 separate d t s e caches n z 2 i i 2 instances l r a e shared caches m p 1.5 r 4 instances o n 1 0.5 0 hmmer libquantum 20 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Resource Contention SPEC CPU 2006 4.5 4 3.5 e m i 1 instance e t 3 n c u 2 instances n r a 2.5 separate d t s e caches n z 2 i i 2 instances l r a e shared caches m p 1.5 r 4 instances o n 1 0.5 0 hmmer libquantum 21 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Resource Contention SPEC CPU 2006 3.5 1 instance 3 2 instances separate caches 2.5 2 instances shared caches 2 4 instances 1.5 1 0.5 0 x s d y r s f g k I o p 2 r k d s 3 p f D c m m I e c e a e a l e x c s n m t p m 3 p l m m a T i r n t m u b m v n l a r e m 4 e i e t s v e p D z a b o b e t l s a i m m 6 j a i n o h o m d b s l n o t u c n F w s 2 a p p s o a g n e m h s e h b u s r g z a m l g o q l a e b x G i l compute-bound memory-bound 22 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Resource Contention SPEC CPU 2006 Compute-bound benchmarks Little resource contention Memory-bound benchmarks Severe slowdown caused by memory contention Huge increase in memory demands since SPEC 2000 Cache contention of comparatively little importance 23 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
Energy Efficiency under DVFS Comparison of 1.6GHz to 2.4GHz 4 instances of benchmark Reducing the frequency pays off for memory intensive tasks 1.8 time 1.6 energy 1.4 edp 1.2 1 0.8 0.6 0.4 0.2 0 d y r f g I o p 2 r d 3 p f x D s s k k s c m m I e e a c a l x e c s n m t p m 3 e p l m m a T i r n m t u b m n a r e v l m e i e t 4 s e p D v z b o b e a s t l a i m m j a i n 6 o m d b h o s n o t u c l n F w s 2 a p p s o a n g e h m s e h b u s r g z a m l g o q l a b e x G i l compute-bound memory-bound 24 Resource-Conscious Scheduling for Energy Efficiency on Multicore Processors
0.5 1.5 2.5 3.5 0.2 0.4 0.6 0.8 1.2 1.4 1.6 1.8 0 1 2 3 0 1 g r o m a c s g r o m a c s 4 instances caches 2 instances shared caches 2 instances separate 1 instance n a m d n a m d p o v r a y p o v r a y h m m e r h m m e r g a m e s s g a m e s s h 2 6 4 r e f h 2 6 4 r e f s j e n g s j e n g g o b m k g o b m k contention d e a l I I d e a l I I DVFS t o n t o t o n t o z e u s m p z e u s m p b z i p 2 b z i p 2 a s t a r a s t a r x a l a n c b m k x a l a n c b m k l e s l i e 3 d l e s l i e 3 d b w a v e s b w a v e s s p h i n x 3 s p h i n x 3 o m n e t p p o m n e t p p m c f m c f s o p l e x s o p l e x G e m s F D T D G e m s F D T D m i l c m i l c edp energy time l i b q u a n t u m l i b q u a n t u m l b m l b m
0 1 2 3 4 5 6 7 8 9 26 Contention Energy Efficiency under DVFS and Resource g r o m a c s n a m d p o v r a y Multicore Processors Resource-Conscious Scheduling for Energy Efficiency on h m m e r g a m e s s c a l c u l i x h 2 6 4 r e f s j e n g p e r l b e n c h g o b m k c a c t u s A D M d e a l I I t o n t o z e u s m p w r f b z i p 2 a s t a r x a l a n c b m k l e s l i e 3 d b w a v e s s p h i n x 3 o m n e t p p m c f s o p l e x G e m s F D T D EDP energy time g c c m i l c l i b q u a n t u m l b m
Recommend
More recommend