Presentation is loading. Please wait.

Presentation is loading. Please wait.

International Conference on Autonomic Computing 2005 1 Governor: Autonomic Throttling for Aggressive Idle Resource Scavenging Jonathan Strickland (1) Vincent.

Similar presentations


Presentation on theme: "International Conference on Autonomic Computing 2005 1 Governor: Autonomic Throttling for Aggressive Idle Resource Scavenging Jonathan Strickland (1) Vincent."— Presentation transcript:

1 International Conference on Autonomic Computing 2005 1 Governor: Autonomic Throttling for Aggressive Idle Resource Scavenging Jonathan Strickland (1) Vincent Freeh (1) Xiaosong Ma (1 & 2) Sudharshan Vazhkudai (2) (1) Department of Computer Science, NC State Univ. (2) Mathematics and Computer Science Division, Oak Ridge National Laboratory

2 International Conference on Autonomic Computing 2005 2 Presentation Roadmap Introduction Model and approach System implementation Performance results Conclusion and future work

3 International Conference on Autonomic Computing 2005 3 Aggregating Desktop Computer Resources Personal computers pervasive  Easily updated and well equipped  Under-utilized Consolidate scattered resources by resource scavenging (resource stealing) Computing resources  Condor, Entropia  SETI@home, Folding@home  Creating massive compute power Storage resources  Farsite, Kosha, FreeLoader  Aggregate distributed spaces into shared storage (Courtesy: SETI@home) (Courtesy: Folding@home)

4 International Conference on Autonomic Computing 2005 4 Impact on Workstation Owners Foremost concern of resource donors  Security and privacy impact Virtual machine/sandbox solutions  Performance impact Existing approaches often too conservative “Stop” approach  Stop scavenging when user activity detected  Unable to utilize small pieces of idle time  Does not overlap scavenging with native workload Priority-based approach  Works for cycle-stealing  Implicit, “best-effort”  Range and granularity limited by operating system

5 International Conference on Autonomic Computing 2005 5 Objectives and Contributions Goal: systematic performance impact control framework Contributions: Governor  Explicit, quantified approach toward performance impact control  Extensible framework for arbitrary scavenging applications and native workloads  User-level, OS-independent implementation  Evaluation with two types of scavenging applications

6 International Conference on Autonomic Computing 2005 6 Presentation Roadmap Introduction Model and approach System implementation Performance results Conclusion and future work

7 International Conference on Autonomic Computing 2005 7 System Entities Active on donated workstations  Resource scavenging application (scavenger)  Native workload  Governor process Controls execution of scavenger Limits impact on native workload to target level α (e.g., 20%)

8 International Conference on Autonomic Computing 2005 8 Performance Impact Performance impact  Caused by resource scavenging application on workstation owner’s native workload  Metrics: slow-down factor (Time scavenged – Time original ) / Time original  May not reflect resource owner perceived impact Main approach: resource throttling  Throttle level (β, 0<=β<1) Time scavenging / Time total  Major challenge: to select appropriate β value

9 International Conference on Autonomic Computing 2005 9 Impact Benchmarking Characterize scavenger S against system resources Native workload as combination of resource consumption components  Resource vector R = (r 1, r 2, …, r n )  Benchmark vector B = (B 1, B 2, …, B n )  Measure S’ impact on B i with various throttle levels Store impact curve  Calculate target throttle level β i with given impact level α

10 International Conference on Autonomic Computing 2005 10 Native Workload Monitoring Native workloads typically complex and dynamic Online workload monitoring  Activate corresponding β when non-trivial native resource consumption detected  Resource trigger vector Т = ( τ 1, τ 2, …, τ n )  For each resource R i β i ’ = 1, 2, … n ) Overall β = min (β 1 ’, β 2 ’, … β n ’ )  Picking most restrictive β across resources β i, if consumption ≥ τ i 1, if consumption < τ i

11 International Conference on Autonomic Computing 2005 11 Governor Architecture scavenger system resources Resource vectors            0. impact benchmarking 1. monitor resource activity 2. compute overall  3. throttle scavenger Governor User target   Adaptive  Extensible and generic

12 International Conference on Autonomic Computing 2005 12 Presentation Roadmap Introduction Model and approach System implementation Performance results Conclusion and future work

13 International Conference on Autonomic Computing 2005 13 Dynamic Throttling Mechanism Fixed throttle interval “I”  1 second in our implementation Within each I, Governor  Runs scavenger application for β*I  Monitors native workload during (1-β)*I  Adjust β for next I 0.2 … Scavenger phases Monitoring phases β=0.5 I β=0.3 I I β=0.6

14 International Conference on Autonomic Computing 2005 14 Resource Usage Monitoring and Triggers At beginning and end of each monitoring phase (1-β)*I  Monitor resource usage CPU: /proc/stat (cycles) Disk: /proc/partitions (blocks) Network: /proc/net/dev (bytes) Triggers ( τ array) Resource Trigger value ( τ ) τ CPU 1% utilization τ IO 0 τ network 0

15 International Conference on Autonomic Computing 2005 15 Presentation Roadmap Introduction Model and approach System implementation Performance results Conclusion and future work

16 International Conference on Autonomic Computing 2005 16 Applications, Benchmarks, and Configurations Scavenger applications  SETI@home Search for signals in slices of radio telescope data Computation-intensive  FreeLoader Prototype for aggregating storage in LAN environments I/O- and network-intensive Single-resource benchmarks  CPU: EP from NAS benchmark suite  I/O: large sequential file read  Network: repeated downloading with wget Linux workstation  2.8GHz Pentium 4, 512MB memory, 80GB disk

17 International Conference on Autonomic Computing 2005 17 Impact Benchmarking Results SETI FreeLoader ResourceImpact level (α) 0.050.100.200.25 β CPU 0.020.050.100.2 β IO 1.0 β network 1.0 ResourceImpact level (α) 0.050.100.200.25 β CPU 0.300.400.700.90 β IO 0.050.100.200.25 β network 0.100.200.300.50

18 International Conference on Autonomic Computing 2005 18 Multi-resource Workload: Kernel Compile Impact on native workload Impact on scavenger app.

19 International Conference on Autonomic Computing 2005 19 Synthetic Composite Workload Simulate common intermittent user activities  Short sleep time between operations Writing 80MB data to file Browsing arbitrary directories in search of file Compressing data written previously and send via networks Browsing more directories Removing files written  Takes about 150 seconds without concurrent user load

20 International Conference on Autonomic Computing 2005 20 Composite Exec. Time and Impact ResourceImpact level (α) 0.00.050.100.200.251.0 SETI@home % impact 142 0% 148 4.0% 154 8.4% 168 18.5% 180 26.8% 261 83.8% FreeLoader % impact 142 0% 150 5.6% 157 10.6% 172 21.1% 180 26.8% 211 48.6% Combine impact benchmarking results with real- time monitoring of composite workload Governor closely approximates target performance impact (α)

21 International Conference on Autonomic Computing 2005 21 Comparison with Priority Based Method (SETI@home)

22 International Conference on Autonomic Computing 2005 22 Comparison with Priority Based Method (FreeLoader)

23 International Conference on Autonomic Computing 2005 23 Presentation Roadmap Introduction Model and approach System implementation Performance results Conclusion and future work

24 International Conference on Autonomic Computing 2005 24 Conclusion and Future Work Governor: extensible framework for quantitative performance impact control  Contains actual performance impact  Proactively consume idle resources  Self-adaptive  OS-independent and low-overhead Future work  Connect impact control with user interfaces  Studying memory resource throttling  Evaluating with more scavengers

25 International Conference on Autonomic Computing 2005 25 Resource Utilization and β for Composite


Download ppt "International Conference on Autonomic Computing 2005 1 Governor: Autonomic Throttling for Aggressive Idle Resource Scavenging Jonathan Strickland (1) Vincent."

Similar presentations


Ads by Google