Soner Yaldiz, Alper Demir, Serdar Tasiran Koç University, Istanbul, Turkey Paolo Ienne, Yusuf Leblebici Swiss Federal Institute of Technology (EPFL), Lausanne, Switzerland Characterizing and Exploiting Task-Load Variability and Correlation for Energy Management in Multi-Core Systems ESTIMedia 2005
2 Multi-Core Soft Real-Time Systems processors Chip-level multiprocessing for massive performance –Energy management problem Real-time multimedia applications –Audio, video processing Soft real-time systems –Tolerance to deadline misses tt + T DEADLINE time start end T2 T4T3 T1 task graph MPEG2 video frames
ESTIMedia Variability and Correlation Capture by Stochastic Models Exploit for Energy Management – –Dynamic Voltage Scaling (DVS) Time Voltage V1 deadline V2 workload Task i variability probability Task 2 workload Task 1 workload positive correlation This work: First approach to consider variability and correlations for multiprocessor energy management
ESTIMedia Application composed of two tasks on a single processor Motivating Example start end T2T1 T DEADLINE = 2 sec Task loads low (2) or high (10) with equal probability Processor Operating Modes – –Slow Mode -> 6 instructions-per-second – –Fast Mode -> 10 instructions-per-second 2 10 instructions T 1,T 2 50% probability
ESTIMedia start end T2T1 T DEADLINE = 2 sec 2 10 instructions T 1,T 2 50% probability T1 T % 10 25% T1 T % % T1 T % 10 50%0 Probabilities for task load combinations: IndependentPositively Correlated Negatively Correlated Task Load Combinations
ESTIMedia T1 T % 10 25% Motivating Example T1 T % % T1 T % 10 50%0 Independent Positively Correlated Negatively Correlated Slow mode -> 12 instructions in 2 sec Misses desired performance never happens ! Fast mode -> 20 instructions in 2 sec Suboptimal energy 1.0 Application – –2 tasks Processor modes – –Slow 6 inst/sec – –Fast 10 inst/sec Deadline – –2 sec Target75% AssumptionIndependent RealityPositive Correlation Target100% AssumptionIndependent RealityNegative Correlation
ESTIMedia Stochastic Modeling Energy Management Scheme –OFFLINE Optimization –ONLINE Adjustments Experimental Results Conclusions OUTLINE
ESTIMedia Stochastic Modeling Flow Computational Demand (CD) of a task –Number of CPU cycles for execution Demands are represented by dist –Quantized for manageability dist is obtained from a set of traces Demand of tasks constitutes an ‘observation’ –(T1,T2) = ( 5, 5 ) observed 3 out of 8. –dist ( 5,5 ) = 3/8 OBSERVATIONS Task1Task start end T2T1 T /8 5 3/8 10 1/8 dist
ESTIMedia MPEG2 video decoding –Widely-used and computationally intensive Slice-based task decomposition(Olukotun et.al,1998) –VLD ( Variable-length decoding) –MC ( Motion compensation ) Case Study: MPEG2 VLD0, MC0 VLD1, MC1 VLD2, MC Experimental Data: – –10 movie segments – –19 slices, 38 tasks – –24 frames-per-second – –~ frames per movie Task Assignment Processor Precedence Data Precedence slice0 slice1slice2
ESTIMedia Variability of MPEG2 Task Loads aggregate one movie aggregate 1- Similarity Traning set predicts workload for others 2- Long Tails Worst-Case causes overdesign one movie
ESTIMedia Correlation among MPEG2 Task Loads High Correlation aggregate statistics one movie Slice 9 Slice 14 Slice 18 Slice 0 Slice 5...
ESTIMedia Critical Path Summation of worst-case task loads : 64 million cycles Observed worst-case total load: 28 million cycles Ignoring correlations lead to far from optimal
ESTIMedia Stochastic Modeling Energy Management Scheme –OFFLINE Optimization –ONLINE Adjustments Experimental Results Conclusions OUTLINE
ESTIMedia OFFLINE: Optimization Formulation Nonlinear constrained optimization problem with 38 variables – –One voltage per task Stochastic programming formulation – –Based on stochastic application model Optimized voltages stored for run-time look-up Each task i has fixed voltage V i for all periods GOAL: Determine optimal V i ’s minimize average energy consumption subject to completion probability
ESTIMedia ONLINE Adjustments When low load is detected, lower the task voltage –Preserving probabilistic performance Very small run-time expense –Few comparisons and arithmetic operations Load lower than expected Slow down further
ESTIMedia Stochastic Modeling Energy Management Scheme –OFFLINE Optimization –ONLINE Adjustments Experimental Results Conclusions OUTLINE
ESTIMedia Experimental Setup Compared with approaches for multiprocessor systems: –I (Zhang et. al, DAC2002 ) Ignores variability, correlations 100% completion Worst-case task load –II ( Hua et. al, EMSOFT2003 ) Ignores correlations Completion Probability Marginal load distribution Training set: 8 movie segments out of 10 Test set has 2 movies not included in training set. Three completion probabilities P CON –0.90, 0.95, 0.99 Two deadlines –Normal, Tight
ESTIMedia Experiment I : Normal Deadline 1. Significant energy savings 2. Desired completion probability achieved Avg E Avg Pr Movie #P CON =0.90P CON =0.95P CON =0.99 IIIOFLNONLNIIIOFLNONLNIIIOFLNONLN
ESTIMedia Experiment II : Tight Deadline Avg E Avg Pr II (Hua2003) fails with tight deadline – –Ignores correlations ONLN improves more Accurate stochastic model
ESTIMedia Experiment III: Comparison with GOD Single MovieOFFLINEONLINEGOD P CON = P CON = P CON = GOD – –Ideal, Unrealizable, Non-causal – –For every individual frame Knows load of each task Computes optimal voltages There is still room for future work – –“application state” structure
ESTIMedia Conclusions Demonstrated significant variability and correlations among workloads of MPEG2 tasks Our stochastic models capture essential characteristics of applications –Accurately predict performance Novel energy management scheme based on stochastic models –Significant energy savings
Soner Yaldiz, Alper Demir, Serdar Tasiran Koç University, Istanbul, Turkey Paolo Ienne, Yusuf Leblebici Swiss Federal Institute of Technology (EPFL), Lausanne, Switzerland Characterizing and Exploiting Task-Load Variability and Correlation for Energy Management in Multi-Core Systems ESTIMedia Questions ?