profiling, prediction, and capping of power in consolidated environments bhuvan urgaonkar computer...
TRANSCRIPT
![Page 1: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/1.jpg)
Profiling, Prediction, and Capping of Power in
Consolidated Environments
Bhuvan UrgaonkarComputer Systems Laboratory
The Penn State University
Talk at Raritan, March 3, 2008
![Page 2: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/2.jpg)
2
Data Center Growth
Explosive growth in both size and numbers Serious implications on
– Robustness of operation• Heterogeneity of hardware/software, workloads, …• Potential lack of scalability of existing resource management solutions• Flash crowds
– Cost of operation• Administrative costs• Power consumption
![Page 3: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/3.jpg)
3
Data Center Growth
Explosive growth in both size and numbers Serious implications on
– Robustness of operation• Heterogeneity of hardware/software, workloads, …• Potential lack of scalability of existing resource management solutions• Flash crowds
– Cost of operation• Administrative costs• Power consumption
![Page 4: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/4.jpg)
4
Growing Power Consumption in Data Centers
Significant energy consumption and growing!– Up to 1.2% of overall power consumption within the US– Growing @ 40% every five years
Growing number of servers main culprit– Increase-per-unit less significant contributor
Lot of interest in dampening this growth
Key technique: Consolidation
![Page 5: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/5.jpg)
5
Consolidation in Data Centers
Goal: Operating/provisioning fewest possible hardware resources while meeting service-level agreements– Our focus: Servers
Multiple spatial scales– Packing multiple applications on a server– Reducing the number of data centers operated by a company
Key ingredients of existing consolidation solutions– Workload characterization and prediction
• Resource requirement inference– Dynamic resource provisioning
• Efficient statistical multiplexing
![Page 6: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/6.jpg)
6
Power Consumption in Consolidated Data Centers
How does consolidation affect power consumption?– How does the power consumed by consolidated aggregates relate to those of
individuals?• Spatial
– Applications, servers, racks, …
• Temporal– Long-term averages: energy consumed, thermal profiles– Short-term peaks: fuses, circuit breakers
– Can we effectively predict these phenomena?• How should we characterize power consumption of individuals?
– Such that we can meaningfully infer behavior upon consolidation
Benefits/utility of such characterization and prediction– Enable consolidation that adheres to “power budgets”– Adapt placement to changes in workloads to obtain desired
performance/power behavior– Determine optimal “power states” (if any) exposed by hardware
• E.g., CPU DVFS states
![Page 7: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/7.jpg)
7
Average and Sustained Power Consumption
Two quantities of interest– Average power consumption– Sustained power consumption
Average and sustained power “budgets” or “caps” of interest at various spatial levels
Our focus: single server consolidating multiple applications
![Page 8: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/8.jpg)
8
Outline
Motivation Power Profiles Power Prediction Preliminary Evaluation
– Power Capping
Ongoing Work
![Page 9: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/9.jpg)
9
Characterizing Power Consumption
Desirable features– Easy/efficient to realize– Amenable to meaningful statistical aggregation
Our approach: Based on offline profiling– Run application in isolation and subject it to realistic workload– Measure power consumption over intervals of chosen length and
construct a PDF• Power profile
![Page 10: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/10.jpg)
10
Offline Profiling Setup
Signametrics SM2040– Measurement rates: 0.2/sec - 1000/sec– Measurement range (AC): 2.5A– Interface: PCI
![Page 11: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/11.jpg)
11
Power Profile Derivation
Power profile: Distribution derived from a representative run
![Page 12: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/12.jpg)
12
Characterizing Power Consumption
Desirable features– Easy/efficient to realize– Amenable to meaningful statistical aggregation
Our approach: Based on offline profiling– Run application in isolation and subject it to realistic workload– Measure power consumption over intervals of chosen length and construct
a PDF• Power profile
Other noteworthy points– Xen-based virtualized hosting
• Each application hosted within a Xen domain
– Easy to do such profiling online– Also measure resource usage and provision enough resources when
consolidating• Appropriate resource managers within the Xen VMM
– Dell PowerEdge server (DVFS-capable CPU)
![Page 13: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/13.jpg)
13
Power Profiles of Real Applications
Applications– SPECjbb– Streaming– SPECInt
• Bzip, MCF
– TPC-W
t = I = 2 msec
![Page 14: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/14.jpg)
14
Variance of Power Profiles
Higher variance (longer tails) for non CPU-saturating applications
Bzip2 Streaming
![Page 15: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/15.jpg)
15
Impact of DVFS State
Non CPU-saturating apps at lower power states
– CPU utilization increases– Power profile less bursty
TPC-W, 60 clients
![Page 16: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/16.jpg)
16
Impact of DVFS State
Power/performance trade-offs depend significantly on how CPU-saturating the application is
TPC-W, 60 clients
![Page 17: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/17.jpg)
17
Outline
Motivation Power Profiles Power Prediction Preliminary Evaluation
– Power Capping
Ongoing Work
![Page 18: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/18.jpg)
18
Average Power Upon Consolidation
Consolidation of CPU saturating applications– Average of individual power consumptions
![Page 19: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/19.jpg)
19
Average Power Upon Consolidation
Consolidation of CPU-saturating and non CPU-saturating– More complex: Some kind of additive effect
![Page 20: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/20.jpg)
20
Average Power: What’s Going On?
CPU+CPU: – Sole significant consumer of power is being time-shared
CPU+non-CPU: – CPU being time-shared
• Though not equally, since non-CPU apps block– CPU and I/O devices being used simultaneously
Insight #1: Separate out power due to resources (such as CPU and I/O)
Insight #2: Also consider the utilization of relevant resources
![Page 21: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/21.jpg)
21
A Simple Predictor for Average Power
Pidle: Power when no application/VM running– Note difference from leakage power
Pbusy/cpu and Pi/o for an application– CPU Power when application running– I/O power due to the application
![Page 22: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/22.jpg)
22
Improved Estimate of Active Power
Capturing non-idle power portion for TPC-W, 60 clients– CPU utilization was 40%
![Page 23: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/23.jpg)
23
Average Power Prediction: Some Results
Prediction accuracy of 2% !– Disclaimer: Pretty small degrees of consolidation
![Page 24: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/24.jpg)
24
Sustained Power Prediction
Goal: Predict the probability that at least S units of power would be consumed for L consecutive seconds
What’s difficult?– Applications that individually do not violate a sustained budget can do so
upon consolidation
![Page 25: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/25.jpg)
25
Sustained Power Prediction
What’s difficult?– Applications that individually do not violate a sustained budget can do so
upon consolidation
time
power
S
L
Violation!
![Page 26: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/26.jpg)
26
Sustained Power Prediction: Some Results
Harder to predict than average– More sophisticated statistical techniques – Omitting details, please find them in our technical
report
Good news: Our profile-based techniques appear to do a good job
![Page 27: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/27.jpg)
27
Example: Power-aware Application Packing
Average budget 180 W Sustained budget 185 W, 1 sec Questions: How many apps can be consolidated and at what
DVFS state?
S1
S2
![Page 28: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/28.jpg)
28
Example: Power-aware Application Packing
Prediction techniques allow systematic answers to previous questions
![Page 29: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/29.jpg)
29
Outline
Motivation Power Profiles Power Prediction Preliminary Evaluation Ongoing Work
![Page 30: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/30.jpg)
30
Ongoing Work
Extending prediction and capping to the rack-level– Employing Raritan’s PDU model DPCR 20-20– Use PDU measurements for online profiling– Prediction of power behavior based on these profiles
Incorporating the storage sub-system’s power consumption– SAN-based array connected to PDU
• Exploring similar profiling techniques
– Flash-based storage
![Page 31: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/31.jpg)
31
Ongoing Work
Efficient provisioning of power infrastructure – Motivation: Recent studies, e.g.,
ISCA paper from Google– Use prediction techniques to
enable closer-to-capacity operation
– Overbook power?– Dynamic adjustment of power
caps• Trade-off energy consumption versus
revenue and thermal constraints
![Page 32: Profiling, Prediction, and Capping of Power in Consolidated Environments Bhuvan Urgaonkar Computer Systems Laboratory The Penn State University Talk at](https://reader035.vdocuments.site/reader035/viewer/2022062409/5697c02d1a28abf838cd9d41/html5/thumbnails/32.jpg)
32
Thank You!
Questions or comments?
More information:– http://csl.cse.psu.edu