with Different Service Levels and Quality of Service Andrei - PowerPoint PPT Presentation

Scheduling for Cloud Computing with Different Service Levels and Quality of Service Andrei Tchernykh CICESE Research center Centro de Investigación Científica y de Educación Superior de Ensenada, Ensenada, Baja California, México chernykh@cicese.mx http://www.cicese.mx/~chernykh Aussois, France , March 31, 2014

Uwe Schwiegelshohn University of Dortmund, Germany Andrei Tchernykh CICESE Research Center, Mexico IPDPS 2012 , IEEE 26th International Parallel and Distributed Processing Symposium

Reserved Instances Spot-Batch Instances On-Demand Instances r 1 r 2 r 3 d j d j d j 2 1 3 time pay for compute capacity by the hour with no long-term commitments If you have flexibility in when your applications can run one-time payment for each instance you want to reserve Amazon resource

Quality of Service f 1 . p j f 2 . p j p j r j d j 1 d j 2 f 2 =4 : guarantees to deliver at least 25% of computing power f 1 =2: guarantees to deliver at least 50% of computing power The provider guarantees to deliver the requested processing time within a certain time frame: slack or stretch factor f i CICESE Parallel Computing Laboratory 4

Quality of Service  Response time in relation to the requested processing time Deadline Service Level (slack Execution time factor and run time execution price) Profit CICESE Parallel Computing Laboratory 5

Scheduling P 1,2,3 =4 f =3 2 4 6 8 10 12 f =3 Optimal 2 4 6 8 10 12 P 4 =1 r 1 =r 2 =r 3 =r 4 =0 d 1 =3 d 2 =d 3 =d 4 =12 CICESE Parallel Computing Laboratory 6

Competitive Factor Obtained Income Competitive Optimal income Factor 𝑜 𝑠 𝑊 ∗ ≥ 𝑊 ∗ = 𝑣 𝑛𝑏𝑦 ∙ 𝑛𝑗𝑜 𝑞 𝑘 , 𝑒 𝑛𝑏𝑦 ∙ 𝑛 𝑘=1 CICESE Parallel Computing Laboratory 7

Scenarios SSL-SM SSL-MM MSL-SM MSL-MM Number of different service levels: single service level or multiple service levels Number of machines: single machine model or multiple machine model CICESE Parallel Computing Laboratory 8

Approach Competitive analysis Worst case ratio between the obtained income and the optimal 𝝇 income over all instances: competitive factor Preemptive Earliest Deadline First scheduling First Fit allocation Greedy acceptance: - a job is accepted if possible. Machine eligibility: - not all jobs are allowed to be executed to every machine. Batch jobs (interactive jobs are not considered) CICESE Parallel Computing Laboratory 9

Competitive Factor 𝒒 𝒏𝒋𝒐 𝟐 𝝇 ≤ 𝟐 − (𝟐 − SSL-SM 𝒒 𝒏𝒃𝒚 ) 𝒈 Das Gupta and Palis, 2001 𝒈 𝝇 ≤ 𝟐 + 𝒈(𝟐 − 𝒒 𝒏𝒋𝒐 SSL-MM 𝒒 𝒏𝒃𝒚 ) CICESE Parallel Computing Laboratory 10

Competitive Factor 𝒒 𝒏𝒋𝒐 𝒈 𝑱 − 𝟐 + 𝒒 𝒏𝒋𝒐 𝒒 𝒏𝒃𝒚 𝒒 𝒏𝒃𝒚 𝝇 ≤ 𝒏𝒃𝒚{ 𝒈 𝑱 − 𝟐 , MSL-SM 𝒈 𝑱 − 𝟐 + 𝒗 𝑱 𝒗 𝑱𝑱 𝝇 ≤ 𝒗 𝑱𝑱 (𝟐 − 𝟐 ) MSL-MM 𝒗 𝑱 𝒈 𝑱 CICESE Parallel Computing Laboratory 11

Energy-Aware Online Scheduling: Ensuring Quality of Service for IaaS Clouds Andrei Tchernykh CICESE Research Center, Mexico Luz Lozano Uwe Schwiegelshohn University of Dortmund, Germany Johnatan Pecero University of Luxembourg, Luxembourg Pascal Bouvry Sergio Nesmachnov Universidad de la República, Uruguay

Model extention • 𝑛 heterogeneous machines 𝑁 𝑗 = (𝑡 𝑗 , 𝑓𝑔𝑔 𝑗 ) – 𝑡 𝑗 speed factor – 𝑓𝑔𝑔 𝑗 energy efficiency 𝑗 = 𝑞 𝑘 /𝑡 𝑗 • 𝑞 𝑘 processing time 𝑗 = 𝑡 𝑗 𝑔 𝑡 𝑘 𝑡 𝑗 is instruction execution speed 𝑡 𝑘 is execution speed guaranteed by SLA CICESE Parallel Computing Laboratory 13

Power consumption 𝐹 𝑝𝑞 = 𝑝 𝑗 𝑢 ∙ 𝑄 𝑗𝑒𝑚𝑓 + 𝑥 𝑗 𝑢 𝑄 𝑥𝑝𝑠𝑙 𝐷 𝑛𝑏𝑦 𝒏 , 𝑗=1 𝑢=1 𝑝 𝑗 𝑢 = 1 , if on ; 0 if off . 𝑥 𝑗 𝑢 = 1 if work , 0 if idle . CICESE Parallel Computing Laboratory 14

Workload Power Energy consumption Proc efficien Speed #U Site Log #Jobs W s cy GFLOPS ser MFS/W 𝑸 𝑱𝒆𝒎𝒇 𝑸 𝑿𝒑𝒔𝒍𝒋𝒐𝒉 DAS2 — University of Amsterdam 64 17.8 35.35 1.36 126 DAS2 — Delft University of Technology 64 17.8 35.35 1.36 126 Gwa-t-1-anon_jobs- DAS2 — Utrecht University 64 17.8 35.35 1.36 126 1124772 333 reduced.swf DAS2 — Leiden University 64 17.8 35.35 1.36 126 KTH — Swedish Royal Institute of Technology 100 17.8 26 1.75 18.6 DAS2 — Vrije University Amsterdam 144 17.8 35.35 1.32 230 KTH-SP2-1996-2.swf 28489 204 HPC2N — High Perf. Comp. Center North, 240 58.9 66 0.89 481 HPC2N-2002-1.1-cln.swf 527371 256 Sweden 88.4 CTC-SP2-1996-2.1- CTC — Cornell Theory Center 430 17.8 26 1.64 79302 679 cln.swf 65.4 LANL-CM5-1994-3.1- LANL — Los Alamos National Lab 1024 24.7 31 1.45 201387 211 cln.swf speed, energy efficiency and power consumption of the machines and their workloads. range of the speed is [18.6, 481] efficiency [0.89, 1.75], power consumption [17.8, 58.9] for 𝑄 𝑗𝑒𝑚𝑓 , and [26, 66] for 𝑄 𝑥𝑝𝑠𝑙 . CICESE Parallel Computing Laboratory 15

Analysis Degradation in performance (relative error) of each strategy under each metric. This is done relative to the best performing strategy for the metric, as follows: 𝑡𝑢𝑠𝑏𝑢𝑓𝑕𝑧 𝑛𝑓𝑢𝑠𝑗𝑑 𝑤𝑏𝑚𝑣𝑓 (𝛿 − 1) ∙ 100 , with 𝛿 = 𝑐𝑓𝑡𝑢 𝑔𝑝𝑣𝑜𝑒 𝑛𝑓𝑢𝑠𝑗𝑑 𝑤𝑏𝑚𝑣𝑓 . Performance profile . 𝜀(𝜐) is a non-decreasing, piecewise constant function that presents the probability that a ratio is within a 𝜐 factor of the best ratio. The function 𝜀(𝜐) is the cumulative distribution function. Strategies with large probability 𝜀(𝜐) for smaller 𝜐 will be preferred. CICESE Parallel Computing Laboratory 16

Strategies Type Strategy Level Description Knowledge-free Allocates job 𝑘 to a machine randomly Rand 1 Allocates job 𝑘 to the first machine available and capable to execute it. Ffit 1 Allocates job 𝑘 to the machine with the least load at time 𝑠 𝑘 : min 𝑗 =1…𝑛 𝑏 𝑜 𝑗 , MLp 1 Allocates job 𝑘 to the machine with higher energy efficiency max 𝑗 =1…𝑛 𝑏 𝑓𝑔𝑔 Max_eff 2 𝑗 Allocates job j to the machine with minimum total power consumption at Energy-aware 𝑠 𝑘 Min_e 2 𝑝𝑞 (𝑢) 𝑗 =1…𝑛 𝑏 time 𝑠 𝑘 : min 𝑄 𝑢=1 𝑗 Allocates job 𝑘 to the machine with the earliest completion time on an 𝑗 𝐷 𝑛𝑏𝑦 𝑗 and 𝑑 𝑙 𝑗 being 𝑗 energy efficient machine 𝑛𝑗𝑜 𝑓𝑔𝑔 𝑗 , where 𝑑 𝑛𝑏𝑦 = max 𝑕 𝑙 =𝑗 𝑑 𝑙 MCT_eff 2 the makespan and completion time of job 𝑙 in the machine 𝑗, respectively Allocates job j to the maximum energy efficient faster machine: Speed- aware 2 𝑗 =1…𝑛 𝑏 𝑡 𝑗 ∗ 𝑓𝑔𝑔 max Max_seff 𝑗 Allocates job j to the fastest machine: max 𝑗 =1…𝑛 𝑏 𝑡 𝑗 Max_s 2 CICESE Parallel Computing Laboratory 17

Average income per machine SLA with 4 SLs FFit Max_eff Max_s Max_seff MCT_eff Min_e MLp Random 4.E+08 4.E+08 3.E+08 income units 3.E+08 2.E+08 2.E+08 1.E+08 5.E+07 0.E+00 1 2 4 8 16 32 64 128 machines CICESE Parallel Computing Laboratory 18

Total income, 4 SLs FFit Max_eff Max_s Max_seff MCT_eff Min_e MLp Random 3.E+09 2.E+09 income units 2.E+09 1.E+09 5.E+08 0.E+00 1 2 4 8 16 32 64 128 machines CICESE Parallel Computing Laboratory 19

Income degradation, 4 SLs Ffit Max_s Max_eff Max_seff MCT_eff Min_e MLp Random 0.9 0.8 0.7 degradation 0.6 0.5 0.4 0.3 0.2 0.1 0 1 2 4 8 16 32 64 128 machines CICESE Parallel Computing Laboratory 20

Total power consumption, 4 SLs FFit Max_eff Max_s Max_seff MCT_eff Min_e MLp Random 3.E+06 2.E+06 2.E+06 Wh 1.E+06 5.E+05 0.E+00 1 2 4 8 16 32 64 128 machines CICESE Parallel Computing Laboratory 21

Degradation of the power consumption Ffit Max_s Max_eff Max_seff MCT_eff Min_e MLp Random 6 5 degradation 4 3 2 1 0 1 2 4 8 16 32 64 128 machines CICESE Parallel Computing Laboratory 22

Performance profile of the income, 8 strategies Random MLp Min_e MCT_eff Max_eff Max_seff Max_s Ffit 100% 90% 80% 70% 60% δ 50% 40% 30% 20% 10% 0% 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 𝜐 CICESE Parallel Computing Laboratory 23

Performance profile of power consumption, 8 strategies Random MLp Min_e MCT_eff Max_eff Max_seff Max_s Ffit 100% 90% 80% 70% 60% δ 50% 40% 30% 20% 10% 0% 1 3 5 7 9 𝜐 CICESE Parallel Computing Laboratory 24

Solution space 1.8 1.6 1.4 Income degradation 1.2 1 0.8 0.6 0.4 0.2 0 0 1 2 3 4 5 6 7 8 9 Power consumption degradation CICESE Parallel Computing Laboratory 25

Pareto 1.80 Random 1.60 Income degradation MLp 1.40 Max_s 1.20 Max_eff 1.00 0.80 MCT_eff 0.60 Min_e 0.40 Max_seff 0.20 FFit 0.00 0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60 1.80 Power consumption degradation CICESE Parallel Computing Laboratory 26

Scheduling for Cloud Computing with Different Service Levels and Quality of Service Andrei Tchernykh CICESE Research center Centro de Investigación Científica y de Educación Superior de Ensenada, Ensenada, Baja California, México chernykh@cicese.mx http://www.cicese.mx/~chernykh Aussois, France , March 31, 2014