Elastic Processes and the Vienna Elastic Computing Model

Transcription

1 Advanced Topics in Service-Oriented Computing and Cloud Computing, the Vienna PhD School of Informatics, WS Elastic Processes and the Vienna Elastic Computing Model Hong-Linh Truong and Schahram Dustdar Distributed Systems Group Vienna University of Technology SOCloud WS

2 Goals Understand different types of elasticity Understand principles of elastic processes Capture the idea of the Vienna Elastic Computing Model SOCloud WS

3 What do we know about elasticity? SOCloud WS

4 Elasticity in physics elasticity (or stretchiness) is the physical property of a material that returns to its original shape after the stress (e.g. external forces) that made it deform or distort is removed It is related to the form (the structure) of something Stress causes the elasticity (structure deformation) Strain measures what has been changed (amount of deformation) In the context of computing: give a process or a system What can be used to represent Stress and Strain? When a strain signals a dangerous situation? How to be elastic under dynamic stress? SOCloud WS

5 Elasticity in economics elasticity is the measurement of how changing one economic variable affects others price elasticity of demand price elasticity of supply income elasticity of demand elasticity of substitution between factors of production elasticity of intertemporal substitution Elasticity of a function: elasticity of y with respect to x SOCloud WS

6 Elasticity in computing (1) Elastic computing is the use of computer resources which vary dynamically to meet a variable workload Clustering elasticity is the ease of adding or removing nodes from the distributed data store What elasticity means to cloud users is that they should design their applications to scale their resource requirements up and down whenever possible., David Chiu Source: SOCloud WS

7 Elasticity in computing (2) About the use of resources in terms of machine powers Mainly for improving the (time-based) performance Reflect a simple physics view of the elasticity Stress is workload Strain is the amount of resources to be increased/decreased SOCloud WS

8 Elasticity in computing another question Charge the user more/less because I could dynamically increase/reduce the quality of my service Quality can response time, quality of results, etc. Why I am able to this? I have access to a large number of and diverse types of (third-party) resources Resources to be used have diverse quality dimensions SOCloud WS

9 Elasticity in computing a broad view (1) To deal with elastic demands from consumers Even a consumer has different requirements at the same time To provide multiple outputs with different price and quality (output elasticity) To deal with elastic data inputs, e.g., deal with opportunistic data (scaling in/out of data inputs) Computing resources are not input data To deal with elastic pricing and quality models associated resources SOCloud WS

10 Elasticity in computing a broad view (2) The service/process itself is elastic in terms of economic theory The service/process is characterized by its elastic cost and quality published to its consumers The service/process structure is elastic in terms of physics Its structure is elastic: e.g., new fragment can be added SOCloud WS

11 Example of elasticity: cloud environments (1) Multiple realtime data sources, e.g. Stream monitoring sensors in smart environments Earth monitoring, satellite images Social media Web information A set of data processing algorithms Data conversion, data enrichment, data extraction, data mining, domain-specific computational analysis (e.g., CO2 calculation), etc Algorithms can be wrapped into processing elements/workflow activities SOCloud WS

12 Example of elasticity: cloud environments (2) Influence factors for the performance and quality of algorithms Data rates, quality of data sources, underlying systems, etc. The quality of the output of algorithms is varying, depending on algorithms and influence factors Multiple concurrent consumers Dynamic requirements for SLA, quality of outputs, etc. Elastic data processing service provider Need to optimize resource usage, prices, etc., but must meet consumer's requirements SOCloud WS

13 Example of elasticity: workflow provisioning and refinement (1) algo1 algo2 algo3 Clouds One processing workflow for all consumers Scheduling, SLA tradeoffs, and individual algorithm impact on the whole processing chain SOCloud WS influence factors performance quality algo4 algo5 Scheduling problems Workflow optimization issues SLA tradeoffs issues Output: results, accuracy, performance etc. =? Reqs: Consumer 1, Consumer 2 Consumer n

14 Example of elasticity: workflow provisioning and refinement (2) algo1 algo2 algo3 prop1:influence factors performance quality algo4 algo5 Output 1: results, accuracy, performance etc. algo1 algo2 algo3 algo4 propn:influence factors Performance quality algo6 algo5 algo7 algo8 Output n: results, accuracy, performance etc. N separated workflows for M consumers Common activities but different influence factors/expected quality SOCloud WS

15 Example of elasticity: workflow provisioning and refinement (3) algo1 algo2 algo3 prop1n:influence factors performance quality algo4 algo5 algo6 Output 1: results, accuracy, performance etc. Merging common workflow fragments algo5 algo7 algo8 Output n: results, accuracy, performance etc. Merging common workflow fragments Dynamic refinements SOCloud WS

16 Principles of Elastic Processes Schahram Dustdar, Yike Guo, Benjamin Satzger, Hong Linh Truong: Principles of Elastic Processes. IEEE Internet Computing 15(5): (2011) SOCloud WS

17 Basic view Consumer demands output1 Input data single process/service outputn output2 resource resource resource resource Resource as data input (e.g., complementary data) SOCloud WS resource resource Resources as executable hosts (e.g., machine, human, software)

18 Elastic Properties price elasticity (1) Price elasticity describes a resource/service's provision's responsiveness to change in price Price items: investment, provisioning, maintenance, etc. Dynamic pricing models can be built based on elasticity concept e.g., Price elasticity of demands SOCloud WS

19 Elastic Properties price elasticity (2) What if no change in individual resource/service Is a traditional issue in service selection? From a customer view Have similar services but different prices Can reduce the usage price But from a provider view Use different resources to offer price elasticity and quality elasticity Build elastic service/process SOCloud WS

20 Elastic Properties quality elasticity Quality elasticity measures how responsive quality is to a change in resource usage Service's quality improvement should be monotonic to the consumption of resources needed Adding more resources does not automatically increase the quality! Multiple quality dimensions Performance, quality of data, etc. Complex dependencies among quality measurement and price function SOCloud WS

21 Conceptual architecture of elastic process environment SOCloud WS

22 Conceptual model: physical elasticity properties EP must optimize resource usage Dynamic environment with diverse resource types Dynamic resources scaling in/out of dynamic resources (inputs and underlying execution environments) EP can produce multiple outputs For demands with similar requirements scaling in/out of multiple outputs and deformation of the structure of EP SOCloud WS

23 Conceptual model: economic elasticity properties EP's function is a static property Accepts inputs and produces outputs Physical elastic properties of EP lead to EP's economic elastic properties Resource elasticity: several instances of EP can be provided elastic Price elasticity: each instance of EP can have different pricing models Quality elasticity: each instance of EP can offer different quality metrics SOCloud WS

24 Conceptual model: operation and modeling principles (1) Operation principles Monitor, manage and describe dynamic properties Dynamically refine process functions based on quality Determine costs based on multiple resource cost models Provide elasticity across providers An EP can deal with multiple service objectives N concurrent consumers, access markets of M providers, with K different requirements: K <= N SOCloud WS

25 Conceptual model: operation and modeling principles (2) Modeling Overlaying EPs, functional composition, and dynamic property composition Modeling principles EP's function as a static property EP's results based on requirements concerning price and quality modeled as a set of constraints influencing resource elasticity Communication can be based on SOAP/REST Refinement and composition of resource, price and quality at multiple levels: activities, fragments, or the whole EP SOCloud WS

26 Research challenges Specification of constraints and preferences How to enrich processes with constraints and preferences specifying pricing and preferences How to allow relevant stakeholders to control trade-offs Self-describing resources Diverse types of resources: hardware, software and humans Elastic reasoning mechanism How to deal with trade-offs and realtime reasonings Reusability and adaptive execution Resoures and fragments refinements based on price and quality Formalism for elastic process systems SOCloud WS

27 The Vienna Elastic Computing Model (VCM) SOCloud WS

28 Key differences Multi-dimensional elasticity Resource, quality, cost and rights Basic on principles of elastic processes Elasticity in hybrid systems of living and nonliving computing resources Software and human capabilities as computing resources Multi clouds An end-to-end approach For science and business complex applications SOCloud WS

29 Motivations for VCM Not only more machines for increasing the performance Quality of results otherwise resources are wasted Price elasticity : taking opportunities in terms of business but a consequence of the demand for a higher of quality of results Problems cannot be solved by software, even with high prices Humans excel machines in many cases Humans in the loop Simulation evaluation, disaster responses, etc. SOCloud WS

30 Motivation example: discover patterns using crowds Jobs of software-based computing elements (SEs) and human-based computing elements (HEs) must be integrated into a single system Quality constraints are important: not just volunteering work! SOCloud WS

31 Motivation example: quality evaluation in multi-scale simulation You need more processional teams/cliques Wisdom of a few! SOCloud WS

32 Conceptual view of VCM E-science Application Business Application Elastic Service Elastic Alignment IT Elastic Process Elastic Runtime Management IT Elastic Service Virtualization Service modeling and interface Alignment techniques Virtualization techniques Monitoring and analysis techniques Non-functional parameters characterization Programming models Governance and service evolution Software-based Computing Element (SE) Human-based Computing Element (HE) Heavy crowded in cloud computing and social computing Ready to use but in isolation (single cloud and single type of computing elements) SOCloud WS

33 Characterizing human-based computing elements Characterizing software-based computing elements is well-researched HE characteristics Computing capabilities: no precise description found in SE, like memory and storage capabilities, and software features Interface: communication interface, task interface, and inter-he interface Dynamic non-functional parameters: lack metrics for HE Programming models: too simple and not well supported. SOCloud WS

34 Analogy between SE and HE Useful for virtualize SE and HE in a single platform Basic computing processor: CPU/core versus brain System architecture: cluster/smp/etc versus crowds/cliques/individual Interconnect: e.g., ring versus free network Communication protocol: TCP/IP versus s Task model:? Versus? Scheduling:? Versus? SOCloud WS

35 The service model in VCM Traditional software service models: Software service and software service-based Models for HE and SE/HE Individual human-based service or human-provided service (HPS) Human team-based service or SCU (social computing unit) Hybrid composite service: a composite service whose aggregated services are software-based or humanbased. SOCloud WS

36 Integrated system view Discover SEs and HEs based on their capabilities, interfaces and nonfunctional parameters Compare services wrapping to SEs and HEs Compose SEs and HEs SOCloud WS

37 Integrated system view: clouds Clouds of SE, HE and hybrid SE/HE systems SOCloud WS

38 Non-functional parameters Rich parameters for SE/HE The same definition for a parameter Different interpretations for SE or HE e.g., availability SOCloud WS

39 Programming framework Program elastic processes of SE/HE SOCloud WS

40 The list of challenges System software and middleware Elastic service modeling and alignment Sevice evolution Programming language and Tools Dynamic refinement and adaptation Charactering and monitoring non-functional parameters Rewarding mechanisms Governance and compliance SOCloud WS

41 Conclusions Elastic computing is needed in different aspects scaling software, services, and people in the same application But we need more than just resource elasticity Resource, cost, quality, etc. Hybrid systems of SE and HE Multiple clouds SOCloud WS

42 References Julien Perez, C\&\#233;cile Germain-Renaud, Bal\&\#225;zs K\&\#233;gl, and Charles Loomis Responsive elastic computing. In Proceedings of the 6th international conference industry session on Grids meets autonomic computing (GMAC '09). ACM, New York, NY, USA, DOI= / Schahram Dustdar, Yike Guo, Benjamin Satzger, Hong Linh Truong: Principles of Elastic Processes. IEEE Internet Computing 15(5): (2011) Anhai Doan, Raghu Ramakrishnan, and Alon Y. Halevy Crowdsourcing systems on the World-Wide Web. Commun. ACM 54, 4 (April 2011), DOI= / D. Schall, H. L. Truong, and S. Dustdar, Unifying human and software services in web-scale collaborations, IEEE Internet Computing, vol. 12, no. 3, pp , S. Dustdar and K. Bhattacharya, The social compute unit, IEEE Internet Computing, vol. 15, no. 3, pp , A. Marcus, E. Wu, D. Karger, S. Madden, and R. Miller, Human powered Sorts and Joins, ArXiv e-prints, Sep Voigt, S.; Kemper, T.; Riedlinger, T.; Kiefl, R.; Scholte, K.; Mehl, H.;, "Satellite Image Analysis for Disaster and Crisis- Management Support," Geoscience and Remote Sensing, IEEE Transactions on, vol.45, no.6, pp , June 2007 doi: /TGRS Schahram Dustdar and Hong-Linh Truong, The Vienna Elastic Computing Model Virtualizing Software and Human for Elastic Processes in Multiple Clouds, Working paper, Nov SOCloud WS

43 Thanks for your attention! Hong-Linh Truong Distributed Systems Group Vienna University of Technology Austria SOCloud WS