Core Technologies for Computing and Future Internet Prof. Kai Hwang University of outhern California This talk attempts to address a few critical issues on cloud computing technology, environment, and emerging new applications in business, science, and society. What are the desired future Internet architecture and protocols to achieve much greater bandwidth, efficient packet routing, and enhanced security than today s Internet? How to use clouds to boost the global economy? What roles clouds can play with a healthy cloud ecosystem to realize the Internet of Things (IoT), to secure pervasive datacenters, and to make the social networks trustworthy? The Big witch in Early 21 st Century Industries race to build massive datacenters (capacity) High bandwidth network and pervasive connectivity Terabit network in sight Wideband wireless mobility Virtualization help realize the economics of scale 2 1 2 ome IT Trends The Data Deluge is clear trend from Commercial (Amazon, e-commerce), Community (Facebook, earch) to cientific applications What are the Computing ervices? (oftware, Platform, Infrastructure, Data, torage, Compute, ) as a ervice (aa, Paa, Iaa, DaaD, aa, Caa,.) Public Light weight clients from smartphones, tablets to sensors Multicore reawakening parallel computing Exascale initiatives will continue drive to high end with a simulation orientation s with cheaper, greener, easier to use IT for (some) applications New jobs associated with new curricula Clients Manager Private s as a distributed system (classic C courses) Data Analytics (Important theme at C11) Other ervices Govt. ervices /Web cience 3 3 4
Applications cience and Technical Applications Business Applications Consumer/ocial Applications Today s ervices tack Application ervices Platform ervices Compute & torage ervices Co-Location ervices ervices 5 6 Business Potential: a trillion $ business/year by 2020? 2 Trillion 600? Distributed and Computing 30% 120? 15% Kai Hwang, Geoffrey Fox, Jack Dongarra, 2016 2020? published by Morgan Kaufmann, Oct. 2011 (648 pages) 7 8
Computing dealing with too many issues yet to be resolved by ACCTA ecurity Privacy Virtualization Trust Qo ervice Level Agreements Legal & Regulatory Resource Metering Billing Provisioning on Demand Programming Env. & Application Dev. Pricing Utility & Risk calability Reliability Energy Efficiency oftware Eng. Complexity 9 The Google Gmail Example http://www.google.com/green/pd/ google-green-computing.pdf s win by efficient resource use over datacenters Business Type Number of users # servers IT Power per user PUE (Power Usage effectivenes s) Total Power per user Annual Energy per user mall 50 2 8W 2.5 20W 175 kwh Medium 500 2 1.8W 1.8 3.2W 28.4 kwh Large 10000 12 0.54W 1.6 0.9W 7.6 kwh Gmail () < 0.22W 1.16 < 0.25W < 2.2 kwh 10 10 Challenges/Issues in Computing 11 11 What Applications Work better in The? Workflow and ervices Pleasingly parallel applications of all sorts analyzing roughly independent data or spawning independent simulations including Long tail of science Integration of distributed sensor data cience Gateways and portals Commercial and cience Data analytics that can use MapReduce (some of such apps) or its iterative variants (most analytic apps) Data Analysis requirements not well articulated in many fields ee http://www.delsall.org for life sciences 12 12
Example 1: Game built at UC GamePipe Lab, No.1 Game program among the to 10 in the UA Reduction of Gaming Latency and Improvement of the Qo and QoE (Frame Rate) on The Game at UC GamePipe Lab. 2012 (Courtesy of Zhao, Hwang, and Villeta, Feb. 2012 [7]) 13 14 Major Technological Challenges in Developing the Future Internet Three Approaches To Building The Future Internet Architecture Programmable ing Architectures Fusion of The Internet, Mobile and TV s OpenFlow for Programmable Virtual ing (tanford, Princeton, etc., 2008) Named Data ing beyond the TCP/IP Content-Centric ing (CCN) Personalized communication protocols Intelligent Routing and Content Distribution : Named Data ing (HP Lab, UCLA, etc. 2009) New Ideas for ecurity and Privacy Protection ervice migration and disaster recovery Massive data protection beyond encryption ervice-oriented Future Internet Architecture (OFIA) : Chinese Academy of ciences, Institute of Computing Technology (2011) 15 16
OpenFlow Architecture and Protocol enable virtual networking, advanced Forwarding and Programmability OpenFlow for Developing The Future Internet 17 March 19, 2012 Prof. Kai Hwang, UC 18 Conventional TCP/IP Internet Protocols OFIA Architecture and Operations The CCN Approach in Named Data ing 19 20
OFIA : ervice-oriented Future Internet Architecture under Development at CA/ICT Roles in The Future Internet (ource: G. Xie, et al, ervice-oriented Future Internet Architecture (OFIA) IEEE Infocom 2011, hanghai, China, April 2011 [2]) 21 22 Ecosystem Requirements: At the system level, the cloud ecosystem include the cloud platform and infrastructure, resource management policies, etc. At the service level, the LAs, globalized standards, reputation system, billing and accounting system, cloud business models, etc. At the user (client) level, Application programming interfaces (APIs), cloud programming environment, Quality of ervice control, etc. is built on Massive Datacenters Range in size from edge facilities to megascale (100K to 1M servers) Economies of scale Approximate costs for a small size center (1K servers) and a larger, 400K server center. Technology torage Administration Cost in smallsized Data Center $95 per Mbps/ Month $2.20 per GB/ Month ~140 servers/ Administrator Cost in Large Data Center $13 per Mbps/ month $0.40 per GB/ month >1000 ervers/ Administrator Ratio 7.1 5.7 7.1 This data center is 11.5 times the size of a football field 23 24
Ecosystem for Market-Oriented s Autonomic Develop methodologies and tools to automate the process of cloud management in 4 objectives 1. Manage resources to provisioning of service quality assurance and adaptation Resource Power 3. Manage energy consumption under LA constraints Capacity Autonomic Load Balancing (ource: R. Buyya, et al, Market-Oriented Computing: Vision, Hype, and Reality for Delivery IT ervices as Computing Utilities, Proc. of HPCC, ept. 25-27, 2008) 25 2. Automate the configuration process of VMs and virtual clusters Admission Control Reliability 4. Develop fault prediction models for proactive failure management 26 26 MapReduce File/Data Repository Parallelism Instruments Map = (data parallel) computation reading and writing data Reduce = Collective/Consolidation phase e.g. forming multiple global sums as in histogram Communication Reduce Portals MPI or Iterative MapReduce Disks Map 1 Map 2 Map 3 /Users Map Reduce Map Reduce Map Example 2: MapReduce kyline Composition of Web ervices in Inter- Applications 27 (Courtesy of L. Chen, K. Hwang, and J. Wu, Jan. 2011 [6]) 28
Reduction of Web ervice Composition Time from 929 ms to 220 ms using fewer kyline representatives Amazon s Lesson Down for 3 days since 4/22/2011 1000x of businesses went offline. e.g. Pfizer, Netflix, Quora, Foursquare,Reddit,. LA contract 99.95% availability (< 4.5 hour down) 10% penalty, otherwise 29 30 Architecture of The Internet of Things Application Layer Merchandise Tracking Environment Protection Intelligent earch Computing Platform Telemedicine Intelligent Traffic mart Home upport in Internet of Things and ocial Applications 1. mart and pervasive cloud applications for individuals, homes, communities, companies, and governments, etc. 2. Coordinated calendar, itinerary, job management, events, and consumer record management (CRM) services Layer Mobile Telecom The Internet Information 3. Coordinated word processing, on-line presentations, web-based desktops, sharing on-line documents, datasets, photos, video, and databases, content distribution, etc. ensing Layer RFID RFID Label ensor ensor Nodes GP Road Mapper 4. Deploy conventional cluster, grid, P2P, social networking applications in the cloud environments, more cost-effectively. 5. Earthbound applications that demand elasticity and parallelism to avoid large data movement and reduce the storage costs 31 32
Example 3: Internet of Things: ensor Grids A pleasingly parallel example on s ensors as a ervice (aa) A sensor ( Thing ) is any source or sink of time series In the thin client era, smart phones, Kindles, tablets, Kinects, web-cams are sensors Output ensor Robots, distributed instruments such as environmental measures are sensors Web pages, Googledocs, Office 365, WebEx are sensors Ubiquitous Cities/Homes are full of sensors They have IP address on Internet ensors being intrinsically distributed are Grids ensors as a ervice However natural implementation uses clouds to consolidate and control and collaborate with sensors ensors are typically small and have pleasingly parallel cloud implementations A larger sensor ensor Processing as a ervice (MapReduce) 33 33 34 Research on ecurity Trusted Zones for VM Insulation Identity federation Virtual network security Access Mgmt Federate identities with public clouds Control and isolate VM in the virtual infrastructu re egregate and control user access APP O APP O Tenant #2 Virtual Infrastructure APP O APP O Virtual Infrastructure Provider Tenant #1 Insulate infrastructure from Malware, Trojans and cybercriminals Insulate information from other tenants Insulate information from cloud providers employees Anti-malware Cybercrime intelligence trong authentication Data loss prevention Encryption & key mgmt Tokenization Courtesy of Dr. Hai Jin, 2011 35 ecurity Info. & Event Mgmt Physical Infrastructure Enable end to end view of security events and compliance across infrastructures GRC (Courtesy of Dr. L. Nick, EMC 2008) 36
s and Grids/HPC ynchronization/communication Performance Grids > s > HPC ystems s appear to execute effectively Grid workloads but are not easily used for closely coupled HPC applications ervice Oriented Architectures and workflow appear to work similarly in both grids and clouds Assume for immediate future, science supported by a mixture of s data analysis (and pleasingly parallel) Grids/High Throughput ystems (moving to clouds as convenient) upercomputers ( MPI Engines ) going to exascale ervice-oriented Architecture (oa) Raw Data Data Information Knowledge Wisdom Decisions Another Grid Another ervice Another Grid Database ervice Compute ervice ervice Another Grid Discovery ervice torage Discovery Traditional Grid with exposed services ensor or Data Interchange ervice Geoffrey Fox: Grid of various clouds -- from Raw Data to Wisdom. = ensor service, = filter services 37 38 Conclusions : At present, cloud technology is still far from being mature and user-friendly. industralization demands first to upgrade of our computer science/engineering education significantly. Interesting IoT applications must leverage the clouds to store and process massive amounts of data, that are changing rapidly on a daily basis. With storming and distrustful clouds, the IoT will be a just dream which will never become true. s, IoT and social networks are rapidly changing our world, reshaping all human relationships, affecting our daily life, and impacting the global economy and political system. We have to be ready to face the new challenges, like it or not! To Probe Further : 1. K. Hwang, G. Fox, and J. Dongarra, Distributed and Computing, Morgan Kauffmann Publishers, Oct. 2011 2. N. Mckeown, et al OpenFlow: Enabling Innovation in Campus s, IGCOMM CCR, Vol. 38, No.2 pp.69-74, 2008 3. G. Xie, et al. ervice-oriented Future Internet Architecture (OFIA) IEEE Infocom, hanghai, China, April 2011 4. R. Buyya, C.. Yeo, and. Venugopal, Market-Oriented Computing: vision, Hype, and Reality for Delivery IT ervices as Computing Utilities, Proc. of IEEE Int l Conf. on HPCC, Dalian, China, ept. 25-27, 2008 5. K. Hwang and D. Li, Trusted Computing with ecure Resources and Data Coloring, IEEE Internet Computing, ept. 2010. 6. L. Chen, K. Hwang, and J. Wu, MapReduce kyline Composition of Web ervices in Inter- Applications, IEEE Trans. Parallel and Distributed ystems, (TPD) submitted under review, 2012. 7. Z. Zhao, K. Hwang and J. Villeta, "GamePipe: A Virtualized Platform Design and Performance Evaluation", ACM Third Workshop on cientific Computing (cience2012), submitted and under reviewing, 2012. 39 40