CSE 691: Energy-Efficient Computing Lecture 6 SHARING: distributed vs. local Anshul Gandhi 1307, CS building

Slides:



Advertisements
Similar presentations
High Performing Cache Hierarchies for Server Workloads
Advertisements

SLA-Oriented Resource Provisioning for Cloud Computing
CSE 691: Energy-Efficient Computing Lecture 20 review Anshul Gandhi 1307, CS building
CPU Scheduling Questions answered in this lecture: What is scheduling vs. allocation? What is preemptive vs. non-preemptive scheduling? What are FCFS,
Energy-efficient Virtual Machine Provision Algorithms for Cloud System Ching-Chi Lin Institute of Information Science, Academia Sinica Department of Computer.
Anshul Gandhi (Carnegie Mellon University) Varun Gupta (CMU), Mor Harchol-Balter (CMU) Michael Kozuch (Intel, Pittsburgh)
CSE 691: Energy-Efficient Computing Lecture 4 SCALING: stateless vs. stateful Anshul Gandhi 1307, CS building
Providing Performance Guarantees for Cloud Applications Anshul Gandhi IBM T. J. Watson Research Center Stony Brook University 1 Parijat Dube, Alexei Karve,
DotSlash – A Web Hotspot Rescue System Weibin Zhao Henning Schulzrinne Department of Computer Science Columbia University June 11, 2004.
Datacenter Power State-of-the-Art Randy H. Katz University of California, Berkeley LoCal 0 th Retreat “Energy permits things to exist; information, to.
Handling Web Hotspots at Dynamic Content Web Sites Using DotSlash Weibin Zhao Henning Schulzrinne Columbia University NYMAN’04.
Handling Web Hotspots at Dynamic Content Web Sites Using DotSlash Weibin Zhao Henning Schulzrinne Columbia University Dagstuhl.
Proteus: Power Proportional Memory Cache Cluster in Data Centers Shen Li, Shiguang Wang, Fan Yang, Shaohan Hu, Fatemeh Saremi, Tarek Abdelzaher.
DotSlash: Providing Dynamic Scalability to Web Applications Weibin Zhao and Henning Schulzrinne Department of Computer Science, Columbia University More.
Power Management in Data Centers: Theory & Practice Mor Harchol-Balter Computer Science Dept Carnegie Mellon University 1 Anshul Gandhi, Sherwin Doroudi,
CSE 691: Energy-Efficient Computing Lecture 3 SLEEP: full-system Anshul Gandhi 1307, CS building
Power Management in Data Centers: Theory & Practice Mor Harchol-Balter Computer Science Dept Carnegie Mellon University 1 Anshul Gandhi, Sherwin Doroudi,
Energy Aware Network Operations Authors: Priya Mahadevan, Puneet Sharma, Sujata Banerjee, Parthasarathy Ranganathan HP Labs IEEE Global Internet Symposium.
GreenHadoop: Leveraging Green Energy in Data-Processing Frameworks Íñigo Goiri, Kien Le, Thu D. Nguyen, Jordi Guitart, Jordi Torres, and Ricardo Bianchini.
Continuous resource monitoring for self-predicting DBMS Dushyanth Narayanan 1 Eno Thereska 2 Anastassia Ailamaki 2 1 Microsoft Research-Cambridge, 2 Carnegie.
Introduction To Windows Azure Cloud
1 An SLA-Oriented Capacity Planning Tool for Streaming Media Services Lucy Cherkasova, Wenting Tang, and Sharad Singhal HPLabs,USA.
Paradigms & Benchmarks Ryan McCune CSE Final Presentation 11/3/11 Notre Dame Computer Science 1.
Computer Science Department University of Pittsburgh 1 Evaluating a DVS Scheme for Real-Time Embedded Systems Ruibin Xu, Daniel Mossé and Rami Melhem.
AUTHORS: STIJN POLFLIET ET. AL. BY: ALI NIKRAVESH Studying Hardware and Software Trade-Offs for a Real-Life Web 2.0 Workload.
Performance Issues in Parallelizing Data-Intensive applications on a Multi-core Cluster Vignesh Ravi and Gagan Agrawal
Bellwether: Surrogate Services for Popular Content Duane Wessels & Ted Hardie NANOG 19 June 12, 2000.
CSE 691: Energy-Efficient Computing Lecture 7 SMARTS: custom-made systems Anshul Gandhi 1307, CS building
CSE 691: Energy-Efficient Computing Lecture 1: Intro and Logistics Anshul Gandhi 1307, CS building
An Operating System Made for E-commerce --On Windows 2000 Advanced Server’s Enterprise-Readiness A Presentation A Presentationfor COSC 513: Operating Systems.
© 2012 IBM Corporation Platform Computing 1 IBM Platform Cluster Manager Data Center Operating System April 2013.
Measuring the Capacity of a Web Server USENIX Sympo. on Internet Tech. and Sys. ‘ Koo-Min Ahn.
DynamicMR: A Dynamic Slot Allocation Optimization Framework for MapReduce Clusters Nanyang Technological University Shanjiang Tang, Bu-Sung Lee, Bingsheng.
1 Adaptive Parallelism for Web Search Myeongjae Jeon Rice University In collaboration with Yuxiong He (MSR), Sameh Elnikety (MSR), Alan L. Cox (Rice),
Kingfisher: A System for Elastic Cost-aware Provisioning in the Cloud
CSE 591: Energy-Efficient Computing Lecture 3 SPEED: processor Anshul Gandhi 347, CS building
CSE 591: Energy-Efficient Computing Lecture 1: Intro and Logistics Anshul Gandhi 347, New CS building
EuroSys Doctoral Workshop 2011 Resource Provisioning of Web Applications in Heterogeneous Cloud Jiang Dejun Supervisor: Guillaume Pierre
CSE 591: Energy-Efficient Computing Lecture 4 SLEEP: full-system Anshul Gandhi 347, CS building
CSE 591: Energy-Efficient Computing Lecture 8 SOURCE: renewables Anshul Gandhi 347, CS building
Power Capping Via Forced Idleness ANSHUL GANDHI Carnegie Mellon Univ. 1.
IIS Progress Report 2016/01/11. Goal Propose an energy-efficient scheduler that minimize the power consumption while providing sufficient computing resources.
CSE 591: Energy-Efficient Computing Lecture 13 SLEEP: sensors
Energy Aware Network Operations
Anshul Gandhi 347, CS building
Anshul Gandhi 347, CS building
Low Power processors in HEP
CSE 591: Energy-Efficient Computing Lecture 17 SCALING: survey
Fair K-Mutual Exclusion Algorithm for Peer to Peer Systems
Scaling the Memory Power Wall with DRAM-Aware Data Management
CSE 591: Energy-Efficient Computing Lecture 20 SPEED: disks
”The Ball” Radical Cloud Resource Consolidation
CSE 591: Energy-Efficient Computing Lecture 21 review
CSE 591: Energy-Efficient Computing Lecture 15 SCALING: storage
Frequency Governors for Cloud Database OLTP Workloads
CSE 591: Energy-Efficient Computing Lecture 10 SLEEP: network
CSE 591: Energy-Efficient Computing Lecture 19 SPEED: memory
HyperLoop: Group-Based NIC Offloading to Accelerate Replicated Transactions in Multi-tenant Storage Systems Daehyeok Kim Amirsaman Memaripour, Anirudh.
CSE 591: Energy-Efficient Computing Lecture 12 SLEEP: memory
CSE 591: Energy-Efficient Computing Lecture 14 SCALING: setup time
CSE 591: Energy-Efficient Computing Lecture 9 SLEEP: processor
Haishan Zhu, Mattan Erez
Zhen Xiao, Qi Chen, and Haipeng Luo May 2013
Energy Efficient Scheduling in IoT Networks
DotSlash: An Automated Web Hotspot Rescue System
Admission Control and Request Scheduling in E-Commerce Web Sites
CSE 531: Performance Analysis of Systems Lecture 4: DTMC
CSE 591: Energy-Efficient Computing Lecture 18 SPEED: power
Take out a piece of paper
Chih-Hsun Chou Daniel Wong Laxmi N. Bhuyan
Presentation transcript:

CSE 691: Energy-Efficient Computing Lecture 6 SHARING: distributed vs. local Anshul Gandhi 1307, CS building

energy_routing paper

# servers

workload

softscale paper

6 Goals of a data center Low response times Goal: T 95 ≤ 500 ms Performance 70% is wasted Goal: Minimize waste Power Load Time  BUSY: 200 W  IDLE: 140 W  OFF: 0 W Intel Xeon server

7 Scalable data centers PerformancePower  BUSY: 200 W  IDLE: 140 W  OFF: 0 W Intel Xeon server Reactive: [Leite’10;Horvath’08;Wang’08] Predictive: [Krioukov’10;Chen’08;Bobroff’07] Setup cost 300 s 200 W (+more) Only if load changes slowly Load Time

8 Problem: Load spikes Load Time x 2x

9 Prior work Dealing with load spikes Spare servers [Shen’11;Chandra’03]  Over provisioning can be expensive Forecasting [Krioukov’10;Padala’09;Lasettre03]  Spikes are often unpredictable Compromise on performance [Urgaonkar’08;Adya’04;Cherkasova’02]  Admission control, request prioritization x Load Time 2x

10 Our approach: SOFTScale No spare servers No forecasting Does not compromise on performance (in most cases) Can be used in conjunction with prior approaches x Load Time 2x

Closer look at data centers Always on Use caching tier to “pick up the slack” 11 Scalable

High-level idea OFF SETUP Load Time x 2x Dual purpose Leverage spare capacity 12 ON

Experimental setup PHP (CPU-bound) Memcached (memory-bound) Response time: Time for entry to exit Average response time: 200ms (with 20X variability) Goal: T 95 ≤ 500ms Apache 13

Experimental setup 14 8-core CPU 4 GB memory 4-core CPU 48 GB memory PHP (CPU-bound) Memcached (memory-bound) Apache

Results: Instantaneous load jumps 15 Load Time 50% 61% 10%  29% T 95 (ms) averaged over 5 mins baseline = provisioned for initial load

Conclusion 16 Problem: How to deal with load spikes? Prior work: Over provision, predict, compromise on performance Our (orthogonal) approach: SOFTScale  Leverages spare capacity in “always on” data tiers  Look at the whole system  Can handle a range of load spikes