Research Challenges in Cloud Computing - Raouf ?· Research Challenges in Cloud Computing ... VL2: A…

Download Research Challenges in Cloud Computing - Raouf ?· Research Challenges in Cloud Computing ... VL2: A…

Post on 03-Jul-2018

212 views

Category:

Documents

0 download

Embed Size (px)

TRANSCRIPT

<ul><li><p>Raouf Boutaba </p><p>Research Challenges in Cloud Computing </p><p>D. Cheriton School of Computer Science University of Waterloo </p><p>CS856 W17 </p></li><li><p>Outline Data Center Networks Network Management Resource and Performance Management Energy Management Pricing and Economics Security and Enterprise Applications </p></li><li><p>Data Center Networks Data center networks form the backbones of data centers Connecting tens of thousands of servers that may host </p><p>millions of applications </p><p>Characteristics Very large scale Single administrative domain Bandwidth is often the performance bottleneck </p><p>3 Research Issues and Current Trends </p></li><li><p>Conventional Architecture </p><p>4 Research Issues and Current Trends Data Center Networks </p><p>Source: VL2: A Scalable and Flexible Data Center Network, SIGCOMM 2009 </p></li><li><p>Limitations of Conventional Architectures </p><p>High oversubscription ratio (i.e. creating bandwidth bottleneck) Typically 1:5, 1:80 or even 1:240 at root </p><p>Poor reliability and utilization </p><p>Static network addresses assignment Fragmentation of resources Difficult to support VM migration due to address </p><p>reconfiguration </p><p>5 Research Issues and Current Trends Data Center Networks </p></li><li><p>Design Objectives Scalability </p><p> Scale to millions of servers without compromising performance </p><p>Economics Built using commodity switches and servers </p><p>Performance Low network diameter Large bisection bandwidth </p><p>Reliability Multiple forwarding paths for host-to-host communication </p><p>Application Support Support address reconfiguration and VM migration </p><p>6 Research Issues and Current Trends Data Center Networks </p></li><li><p>Architectural Proposals Switch-Centric </p><p> Forwarding using only switches E.g. Portland, VL2 </p><p>Server-Centric Forwarding using both switches and servers E.g. DCell, Bcube, CamCube </p><p>7 Research Issues and Current Trends Data Center Networks </p></li><li><p>Portland Uses a fat-tree topology for path diversity and large bisection bandwidth Operates on Layer 2 </p><p> Using Pseudo-MAC address in the format of pod.position.port.vmid for forwarding </p><p> Using a centralized fabric manager to manage actual to pseudo MAC mapping </p><p>8 Research Issues and Current Trends Data Center Networks Switch-Centric </p></li><li><p>BCube Targeting container-based datacenters Using a generic hypercube topology </p><p> Overlay routing at layer 2.5 Efficient support for communication patterns such as one-to-one, one-to-</p><p>many, many-to-many using source routing </p><p>9 Research Issues and Current Trends Data Center Networks Switch-Centric </p><p> BCube0 = n servers + one mini-switch (n</p></li><li><p>Research Challenges Understanding the trade-off between different architectures </p><p> Switch centric vs. Server centric </p><p>Comparison criteria Network capacity Robustness Capital and Operational Cost </p><p>Managing and upgrading existing data center networks over time </p><p>10 Research Issues and Current Trends Data Center Networks </p></li><li><p>Outline Data Center Networks Network Management Resource and Performance Management Energy Management Pricing and Economics Security Management Migrating Enterprise Applications to the Cloud </p><p>11 </p></li><li><p>Network Management Issues Naming and addressing </p><p> Address configuration and management </p><p>Flow control and management Congestion Control Flow Scheduling </p><p>12 Research Issues and Current Trends </p></li><li><p>Address Configuration ID/Locator separation is a design principle of data center </p><p>networks. E.g. Portland maintains the mapping between physical MAC and </p><p>hierarchical PMAC addresses, E.g. BCube assigns virtual addresses to individual host </p><p>Automatic address reconfiguration is a requirement Manual configuration is costly and error prone </p><p>13 Research Issues and Current Trends Network Management </p></li><li><p>Congestion Control Data center traffic typically consists of </p><p> (&gt;80%) Low latency short flows (i.e. user facing requests) (</p></li><li><p>Flow Scheduling Given path diversity provided by data center networks, route </p><p>network flows to minimize congestion </p><p>Current Approaches Equal Cost Multipath (ECMP) </p><p> Determining path using a hash function (called flow-hashing) Valiant Load Balancing (VLB) </p><p> Bouncing packet off of random intermediary nodes (switches or servers) </p><p> Limitation: Inefficient for non-uniform traffic patterns </p><p> Two heavy weight flows may collide, resulting in congestion </p><p>15 Research Issues and Current Trends Network Management </p></li><li><p>Flow Scheduling (cont) Flow scheduling </p><p> Separate flows into large and small flows </p><p>For small flows, use ECMP or VLB </p><p>For large flows, use centralized scheduling A variant of NP-hard multi-commodity flow problem </p><p> Implementation Monitor network flows Dynamically inserting forwarding entries for large flows </p><p>16 Research Issues and Current Trends Network Management </p></li><li><p>Research Directions Configuration Management </p><p> Reducing the complexity of management tasks such as address configuration </p><p>Traffic Management Support various usage patterns of cloud applications </p><p> Leveraging new network management paradigms such as SDN </p><p>17 Research Issues and Current Trends Network Management </p></li><li><p>Outline </p><p> Data Center Networks Network Management Resource and Performance Management Energy Management Pricing and Economics Security Management Migrating Enterprise Applications to the Cloud </p><p>18 </p></li><li><p>Resource and Performance Management A cloud computing environment hosts myriads of </p><p>applications with diverse performance objectives </p><p>How to effectively allocate resources to applications to satisfy their performance objectives? </p><p>Sub-problems Performance modeling and management for each individual </p><p>application Run-time resource management </p><p>19 Research Issues and Current Trends </p></li><li><p>Application Performance Management An application owner needs to understand the performance model of the </p><p>application, and adjust resource requirement according to workload condition E.g. Increase number of web server replicas to mitigate flash crowd effect </p><p>20 Research Issues and Current Trends Resource &amp; Performance Mgmt </p><p>Demand Prediction Controller Application </p><p>Performance Model </p><p>Output </p><p>Input </p></li><li><p>Application Performance Management (cont) </p><p>Using probabilistic / statistical methods Queuing Models Machine learning </p><p>Proactive vs. reactive Control Proactive control uses predicted demand to allocate resources before </p><p>they are needed Reactive control respond to immediate demand fluctuations when </p><p>prediction is not available. </p><p>21 Research Issues and Current Trends Resource &amp; Performance Mgmt </p></li><li><p>Data Center Resource Management Objectives </p><p> Mitigating performance bottleneck (i.e. hotspot) Improving application schedulability Improving server utilization Improve resource sharing among applications Reducing energy cost </p><p>Current approach: using various virtualization techniques Dynamically adjusting resource allocation of applications Virtual machine migration </p><p>22 Research Issues and Current Trends Resource &amp; Performance Mgmt </p></li><li><p>Data Center Resource Management (cont) Optimal placement problem is a general case of multi-</p><p>dimensional bin packing problem NP-hard to solve </p><p>Additional Factors Job arrival process Job duration Reconfiguration procedure and cost </p><p> E.g. cost of migration </p><p>23 Research Issues and Current Trends Resource &amp; Performance Mgmt </p></li><li><p>Research Directions Understanding application resource requirements </p><p> e.g. workload characterization, application performance analysis </p><p>Resource management framework for data-center wide workloads </p><p>Multi-tenancy issues Application owner and cloud owner may have potentially conflicting </p><p>objectives </p><p>24 Research Issues and Current Trends Resource &amp; Performance Mgmt </p></li><li><p>Outline </p><p> Data Center Networks Network Management Resource and Performance Management Energy Management Pricing and Economics Security Management Migrating Enterprise Applications to the Cloud </p><p>25 </p></li><li><p>Energy Management Reducing energy consumption is a critical objective of cloud </p><p>computing </p><p>Power and cooling cost constitutes a large potion of datacenter expenditure 25%-30% total data center operational cost </p><p>Government regulations call for environment friendly (i.e. Green) data centers </p><p>26 Part 2- Research Issues and Current Trends </p></li><li><p>Cost of Consumption Power and Cooling cost millions of dollars monthly </p><p>27 Part 2- Research Issues and Current Trends Energy Management </p><p>Estimated Monthly Operational Expenditure of a 50k machine Data Center Source: http://perspectives.mvdirona.com/ </p></li><li><p>Reducing Energy Cost Server Consolidation </p><p> Reducing number of servers used by turning off unused servers </p><p>Energy-Aware scheduling Scheduling jobs to reduce power and cooling costs </p><p>Energy Efficient Networks Dynamically adjust active network elements to reduce power </p><p>cost </p><p>28 Part 2- Research Issues and Current Trends Energy Management </p></li><li><p>Server Consolidation Consolidating application workloads on a smaller number of </p><p>servers to save server power cost </p><p>However, consolidation increases resource contention among applications, which may hurt their performance </p><p>Challenges Understanding the energy and performance impact of consolidation Devising effective policies for achieving good trade-off between </p><p>performance and power cost </p><p>29 Part 2- Research Issues and Current Trends Energy Management </p></li><li><p>Energy-aware Workload Scheduling Power-aware scheduling </p><p>Schedule jobs to minimize server power consumption E.g. leveraging Dynamic Voltage and Frequency Scaling (DVFS) to </p><p>reduce server power consumption Thermal-aware scheduling </p><p>Scheduling jobs to minimize overall data center temperature E.g. scheduling jobs to reduce server temperature so as </p><p>to reduce cooling cost </p><p>30 Research Issues and Current Trends Energy Management </p></li><li><p>Energy Efficient Networks Objective: making data center networks energy-proportional </p><p> Make energy cost proportional to network utilization </p><p>Approach: Given the current network condition, dynamically adjust active network elements to reduce power cost Powering down unneeded switches and links Adjusting link rate </p><p> Many modern switch models (e.g. infiniBand) can specify more than one operation range </p><p>31 Research Issues and Current Trends Energy Management </p></li><li><p>Research Directions Effectively leveraging latest hardware, software technologies </p><p>to achieve high energy cost reduction </p><p>Achieving a good trade-off between performance and energy cost E.g. Reducing CPU rate using DVFS slows down job execution </p><p>32 Research Issues and Current Trends Energy Management </p></li><li><p>Outline </p><p> Data Center Networks Network Management Resource and Performance Management Energy Management Pricing and Economics Security Management Migrating Enterprise Applications to the Cloud </p><p>33 </p></li><li><p>Pricing and Economics Cloud computing is a realization of utility computing </p><p> Provide storage and computing resources using a usage based pricing model </p><p>Demand is highly volatile in Cloud environments Low resource demand causes low server utilization High resource demand results in unsatisfied demands, which causes </p><p>customer dissatisfaction </p><p>34 Research Issues and Current Trends </p></li><li><p>Pricing and Economics (cont) Approach: using market economy to shape demand </p><p> Dynamically adjust resource supply and price </p><p> Increase price when demand spikes Ensure resources are allocated to most needing users Provide incentive for customers to reduce demand </p><p>Reduce price when demand is low Incentivize customers to increase demand </p><p>35 Part 2- Research Issues and Current Trends </p></li><li><p>Amazon EC2 Spot Market Amazon EC2 launched spot </p><p>instance service in Dec. 2009 </p><p> Price of resources fluctuates with supply and demand Customers specify their bids in </p><p>their resource requests A market-based mechanism </p><p>decides the final price and assign resources to customers </p><p>36 Part 2- Research Issues and Current Trends Pricing and Economics </p><p>Price of m1.small linux spot instance in US-West-1 from Sept. 24-Sept. 30, 2010 </p><p>(Source: www.cloudexchange.org) </p></li><li><p>Market-Oriented Resource Allocation Objectives </p><p> Truthful, fair and revenue maximizing </p><p>Additional considerations Support price discovery </p><p> Providing historical prices Easy to compute </p><p> Solving NP-hard problems in real-time is not preferred </p><p>37 Part 2- Research Issues and Current Trends Pricing and Economics </p></li><li><p>Research Directions Designing and analyzing pricing schemes for cloud </p><p>computing Satisfy all previous objectives is difficult Most of the existing work use auction mechanisms but mostly focus </p><p>on single-round auctions Need to understand dynamics for multi-round repeat auctions </p><p>More general pricing scheme Packaging Volume discount </p><p>38 Research Issues and Current Trends Pricing and Economics </p></li><li><p>Outline </p><p> Data Center Networks Network Management Resource and Performance Management Energy Management Pricing and Economics Security Management Migrating Enterprise Applications to the Cloud </p><p>39 </p></li><li><p>Security Management Cloud customers are concerned about privacy and </p><p>confidentiality of their data and applications in the Cloud </p><p>Security risks Information leaking and stealing by </p><p> Adversarial users in the cloud Cloud providers </p><p> Attacks within data centers Performance interference and disruption Denial of Service (DoS) attack </p><p>40 Research Issues and Current Trends </p></li><li><p>Security Management (cont) Security in traditional environments </p><p> Application owners can modify the security settings of the underlying fabric </p><p>Security in cloud computing environment Underlying fabric is operated by the cloud infrastructure provider Individual application owners cannot directly modify security settings Different stakeholders may have potentially conflicting interests </p><p>41 Research Issues and Current Trends </p></li><li><p>Security in the Cloud Establishing trust between Cloud providers and customers </p><p> Cloud provider continuously monitor and audit customers VMs Customer Privacy enforcement through attestation </p><p> Relying on trusted platform module (TPM) Using non-forgeable hardware signatures to prove no non-privileged </p><p>memory access has been done </p><p>Auditability must be mutual between providers and customers Since both sides can be malicious </p><p>42 Research Issues and Current Trends Security Management </p></li><li><p>Research Directions Supporting fine grained security requirements </p><p>Different users will have different security needs </p><p>Eliminating source of information leakage E.g. side channels through memory cache </p><p>Minimizing impact of auditing on performance </p><p>43 Research Issues and Current Trends Security Management </p></li><li><p>Outline </p><p> Data Center Networks Network Management Resource and Performance Management Energy Management Pricing and Economics Security Management Migrating Enterprise Applications to the Cloud </p><p>44 </p></li><li><p>Migration of Enterprise Applications Outsourcing (or partially outsourcing) enterprise </p><p>infrastructure to the cloud is a growing trend in the industry Reducing capital investment and maintenance cost </p><p>Challenges Find a cost-effective strategy for outsourcing Integration with existing business infrastructure Security and privacy </p><p>45 Research Issues and Current Trends </p></li><li><p>Research Directions Selecting cloud services among multiple providers </p><p> Evaluating service offerings in terms of performance, reliabilit...</p></li></ul>