a broker for cost-efficient qos aware resource allocation in ec2. kurt vermeersch coordinator: kurt...
TRANSCRIPT
![Page 1: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/1.jpg)
A Broker for Cost-efficient QoS aware Resource Allocation in EC2.
Kurt Vermeersch
Coordinator: Kurt Vanmechelen
![Page 2: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/2.jpg)
2
Thesis Goal
Mapping of QoS constrained workloads to Amazon pricing models, while trying to
minimize the total cost.
Broker Seller <-> Buyer
Cost-efficient Minimize cost
QoS aware Constraints
Resource Allocation Scheduler
EC2 Amazon
![Page 3: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/3.jpg)
3
Cloud Computing [1/2]
“Cloud computing is a large-scale distributed computing paradigm that is driven by economies of scale, in which a pool of abstracted, virtualized, dynamically-scalable, managed computing power, storage, platforms, and services are delivered on demand to external customers over the Internet.”
Ian Foster, Cloud Computing and Grid Computing 360-Degree Compared
![Page 4: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/4.jpg)
4
Cloud Computing [2/2]
• Distributed- location and device independence
• Economies of Scale - less expensive resources
• Virtualized - server consolidation
• Dynamically-scalable - no over or under provisioning- illusion of infinite amount- capEx to opEx
![Page 5: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/5.jpg)
5
Amazon Cloud Computing
• Why?- Knowledge- Diversification
• What?- Public : an off-site third-party cloud provider- IaaS: hardware resources are provided- Product Portfolio: EC2, S3, etc.- ECU: equivalent CPU capacity of a 1.0-1.2 GHz
2007 Opteron
![Page 6: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/6.jpg)
6
Amazon Instance Types
![Page 7: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/7.jpg)
7
Amazon Pricing• Four Regions
- US East, US West, EU and Asia Pacific
• On-Demand- Fixed hourly charging rate - Guaranteed to stay alive
• Reserved- Upfront payment (1y/3y), lower fixed hourly rate- Guaranteed to be available for launch
• Spot- Varying hourly rate- Instance can be terminated
![Page 8: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/8.jpg)
8
Environmental Analysis
Variation in products, instances and pricing models complicates the mapping from workload to optimal resource division
Pricing Analysis Heuristic
Workload Analysis BrokerEtc.
![Page 9: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/9.jpg)
9
On-Demand Pricing Evolution
![Page 10: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/10.jpg)
10
On-Demand Region Comparison
![Page 11: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/11.jpg)
11
Reserved Pricing Evolution
![Page 12: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/12.jpg)
12
Spot Analysis
• Spot price history => CSV files- Cloudexchange.org- EC2 API
• Statistical analysis of Spot price evolution- High Memory Extra Large in US-East
• Analysis using average Spot prices
![Page 13: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/13.jpg)
13
Spot History [1/4]
• fluctuating average price
![Page 14: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/14.jpg)
14
Spot History [2/4]
• Aligned boxplot percentiles => outliers
![Page 15: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/15.jpg)
15
Spot History [3/4]
• Only small differences during the day
![Page 16: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/16.jpg)
16
Spot History [4/4]
• Only a little cheaper during the weekend
![Page 17: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/17.jpg)
17
Spot Average [1/3]
![Page 18: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/18.jpg)
18
Spot Average - Regions [2/3]
![Page 19: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/19.jpg)
19
Spot Average – Regions [3/3]
![Page 20: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/20.jpg)
20
Broker Design
![Page 21: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/21.jpg)
21
Workload & Constraints
W1: total VM W2: Every hourhours specified # VMs specified
![Page 22: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/22.jpg)
22
Spot Decision Model
• Based on empirical data• Checkpointing schemes
- Hourly- Optimal
• Decision model determines spot bid that minimizes cost but ensures successful execution in terms of workload constraints- Java port: memory problem fix
![Page 23: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/23.jpg)
23
Distributor
• Make division between different pricing models after scheduling has occurred- On-Demand vs Reserved -> optimal division possible- Spot vs Reserved -> optimal choice spot
• Spot is not always the best choice cfr. constraints.
![Page 24: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/24.jpg)
24
On-Demand vs Reserved [1/3]
![Page 25: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/25.jpg)
25
On-Demand vs Reserved [2/3]
![Page 26: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/26.jpg)
26
On-Demand vs Reserved [3/3]
![Page 27: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/27.jpg)
27
Scheduler
• Capacity fragmentation • Based on workload model• Computation intensive
• Basic scheduling vs Optimized scheduling
![Page 28: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/28.jpg)
28
Basic Scheduling (w1)
![Page 29: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/29.jpg)
29
Optimized Scheduling (w1) [1/2]
edfSort(tasks);for(task t : tasks){
for(instance i : instances){i.addPartTillDeadlineOrEnd(t);if(t.isDistributed()){ break; }if(i.isLast()) { instances.addNew(); }
}}
![Page 30: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/30.jpg)
30
Optimized Scheduling (w1) [2/2]
![Page 31: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/31.jpg)
31
Basic Scheduling (w2)
![Page 32: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/32.jpg)
32
Optimized Scheduling (w2) [1/2]
for(Task t : tasks){buckets.divideEqually(t);
}for(Bucket b : buckets){
//try all combinations, choose the one//that minimizes the number //of needed instancesb.makePlanning();
}
![Page 33: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/33.jpg)
33
Optimized Scheduling (w2) [2/2]
![Page 34: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/34.jpg)
34
What’s next?
• Extend scheduler with a checkpointing cost• Extend scheduler/broker with spot instances
- Using findings from spot analysis- Using decision model software
• Evaluate cost cuttings achieved by broker• A lot of writing!
![Page 35: A Broker for Cost-efficient QoS aware Resource Allocation in EC2. Kurt Vermeersch Coordinator: Kurt Vanmechelen](https://reader035.vdocument.in/reader035/viewer/2022062518/56649e9c5503460f94b9d1b1/html5/thumbnails/35.jpg)
35
Thank You!
Questions?
Check out http://www.thesis.kurtvermeersch.com