Software development methodology in a Green IT environment

Transcription

1 Software development methodology in a Green IT environment Hayri ACAR a, Gülfem I. ALPTEKIN b, Jean-Patrick GELAS a,c, Parisa GHODOUS a a Université Claude Bernard, Lyon 1, LIRIS, France c ENS Lyon, LIP, UMR 5668, France b Galatasaray University, Turkey Conférence éco-conception logicielle 03/02/ /02/2017 1

2 Usage and users never stop growing Internet live stats 03/02/2017 2

3 INTRODUCTION USAGE: RESULTS: Applications Software Mobile devices (Smartphone, Tablet, ) Web Services Cloud Internet Power Consumption Energy Consumption Greenhouse Gas Emission Pollution How we can optimize our usage to obtain green results? 03/02/2017 3

4 STATE OF THE ART 03/02/2017 4

5 STATE OF THE ART POWER CONSUMING COMPONENTS CPU MEMORY HARD DISK NETWORK OTHERS 03/02/2017 5

6 STATE OF THE ART Software Tools Hybrid solutions Hardware methodologies 03/02/2017 6

7 STATE OF THE ART - Software tools: Based on Mathematical formula to estimate Energy consumption. Very few tools in this area of study. Tools are not exhaustive. 03/02/2017 7

8 SYSTEMATIC REVIEW 03/02/2017 8

9 Some software tools Tools Power Model Appreciations Joulemeter E system Globally, estimates energy = E CPU + E Memory + E Disk consumption of all components, but for a given process estimates only energy consumption of CPU. vec E = E bus + E cell + E pad + E main Estimates only energy consumption due to memory. Orion E = E read + E write Communication components are considered. Span P(t j, f i ) pret = P(t j,f i ) pret + P(f i ) In the Software code, manually code can be added to show the parts of code involved on the power consumption. PowerAPI P Software = P comp + P com Only CPU and network have been considered. 03/02/2017 9

10 STATE OF THE ART Power saving techniques Transistor sizing and reordering Network-On-Chip Reduce memory access Dynamic power consumption of transistors is decreased Improvement over standard bus and crossbar interconnections integrate a cache in the classical memory hierarchy of a modern processor DFS (Dynamic Frequency Scaling) Method where frequency of a microprocessor can be adapted automatically DVS (Dynamic voltage scaling) Lowering or increasing the supply voltage of the CPU 03/02/

11 ENGINEERING PROCESS Green analysis Green analysis Requirements Design & Implementation Tests Disposal Maintenance Usage Tow ards a Green and Sustainable Software H. Acar, G.I. Alptekin, J.-P. Gelas and P. Ghodous. In Proceedings of the 22nd {ISPE Inc. International Conference on Concurrent Engineering. 03/02/

12 POWER MODEL Dynamic and static power: P Software = P dynamic + P static Separating dynamic and static power: P Softwre = P CPU,dynamic + P CPU,static + P Memory,dynamic + P Memory,static + P Disk,dynamic + P Disk,static + P Network,dynamic + P Network,static 03/02/

13 POWER MODEL CPU: P CPU = C V² F P CPU,id = P CPU N id 100 Memory: P DRAM,dyn = P Activate + P Precharge + P Read + P Write P DRAM,id = P DRAM,dyn. M id Hard Disk: P Disk = P Active + (P Idle + P Standby + P Sleep ) P Disk,dyn = P Activate = P Read + P Write 03/02/

14 TOP PROGRAMMING LANGUAGE IEEE Spectrum TIOBE INDEX 03/02/

15 Tool for Estimating Energy Consumption 03/02/

16 EXPERIMENTS Fibonacci sequence is implemented in Java of 2 ways: - Recursively: 45 terms are calculed. - Iteratively: calculation of 5,000 values. TEEC & Joulemeter results are compared. 03/02/

17 EXPERIMENTS Quite similar results between TEEC and Joulemeter. Recursive method consumes more power in amplitude and time than the iterative method. 03/02/

18 EXPERIMENTS Unoptimized code Optimized code Functions Unoptimized Optimized Time (ms) Locality of reference Compare array to array list Compare integer list loop Char array StringBuilder /02/ Binary search

19 EXPERIMENTS 03/02/

20 C EXPERIMENTS Nested Loops Java Unoptimized Optimized Unoptimized Optimized char* fn(int N) { char* fn2(int N) { public static char[] fn(int N) public static char[] fn2(int int i = 0; int i = 0; { N) { char* v = (char*)malloc(n * 2); char* v = (char*)malloc(n * int i = 0; int i = 0; 2); while (i < N) { char[] v = new char[n*2]; char[] v = new char[n*2]; int tmp; int j = 0; while (i < N) { int tmp; v[i] = 0; while (i < N) { int j = 0; while (i < N) { int j = 0; i++; v[i] = 0; int j = 0; v[i] = 0; while (j < N) { i++; v[i] = 0; v[i] += v[j + N]; i++; while (j < N) { i++; tmp = 0; j++; v[i] += v[j + N]; tmp = 0; while (j < N) { for (v + j + N) while (j < N) { tmp += v[j + N]; j++; tmp += v[j + N]; j++; return v; j++; v[i] = tmp; return v; v[i] = (char) tmp; return v; return v; 03/02/

21 EXPERIMENTS 03/02/

22 CONCLUSIONS and FUTURE WORKS - Currently, energy estimation tools aren t exhaustive. - TEEC: now estimates CPU, memory, hard disk and network interface card. - Java agents to determine source code hotpoints. - Propose green programing best practises (with our expertise). - Focus on Cloud computing (web services) 03/02/

23 PUBLICATIONS The impact of source code in software on power consumption H. Acar, G.I. Alptekin, J.-P. Gelas and P. Ghodous. International Journal of Electronic Business Management, TEEC : Improving power consumption estimation of software H. Acar, G.I. Alptekin, J.-P. Gelas and P. Ghodous. In Proceedings of the 30th International Conference on Environmental informatics. Berlin, Germany, September 14-16, Beyond CPU: Considering Memory Power Consumption of Software H. Acar, G.I. Alptekin, J.-P. Gelas and P. Ghodous. In Proceedings of the 5th International Conference on Smart Cities and Green ICT Systems (SMARTGREENS). Rome, Italy, April 23-25, A Green approach to save energy consumed by software H. Acar, G.I. Alptekin, J.-P. Gelas and P. Ghodous. In Proceedings of the 3rd International Conference on ICT for Sustainability (ICT4S). Copenhagen, Denmark, September 7-9, Towards a Green and Sustainable Software H. Acar, G.I. Alptekin, J.-P. Gelas and P. Ghodous. In Proceedings of the 22nd {ISPE Inc. International Conference on Concurrent Engineering. Delft, The Netherlands, July 20-23, 2015, pp /02/