admin

loud computing has gained significant popularity over past few years. Running application on the cloud can benefit from many features of it such as seamless maintenance and high resource availability. Furthermore, employing service­oriented architecture and resource virtualization technology, cloud provides the highest scalability for enterprise applications having variant load. This feature of cloud is the main attraction for migration of workflows to the cloud. A workflow is comprising of several tasks communicating each other through network by sending various data volumes. The communication between tasks in a workflow sometimes turns into main bottleneck in migration of workflows to the cloud. In order to resolve this problem, we partition the workflow and share data files between tasks communicating each other more often. In addition, these tasks perform different operations, requiring different amount of resources. Since each task of a workflow requires different processing power to perform its operation, at time of load variation it must scale in a manner fulfilling its specific requirements the most. Scaling can be done manually, provided that the load change periods are deterministic, or automatically, when there are unpredicted load spikes and slopes in the workload. Enterprise application providers always look for the method that, minimizing resource consumption cost and handling maximum load, guarantees Quality of Service(Qos) expected from them. A number of auto­scaling policies have been proposed so far. Some of these methods try to predict next incoming loads, while others tend to react to the incoming load at its arrival time and change the resource setup based on the real load rate rather than predicted one. However, in both methods there is need for an optimal resource provisioning policy that determines how many servers must be added to or removed from the system in order to fulfill the load while minimizing the cost. Current methods in this field take into account several of related parameters such as incoming workload, CPU usage of servers, network bandwidth, response time, processing power and cost of the servers. Nevertheless, none of them incorporates the life duration of a running server, the metric that can contribute to finding the most optimal policy. This parameter finds importance when the scaling algorithm tries to optimize the cost with employing a spectrum of various instance types featuring different processing powers and costs. We have created a generic LP(linear programming) model that takes into account all major factors involved in scaling including periodic cost, configuration cost and processing power of each instance type, instance count limit of clouds, and also life duration of each instance with customizable level of precision, and outputs an optimal combination of possible instance types suiting each task of a workflow the most. Our goal is to build a framework based on our LP model and workflow partitioning for optimal migration of any enterprise applications/workflows to the cloud.

Research staff

Assoc. Prof. Dr. Satish Narayana Srirama

Alireza Ostovar

Jaagup Viil

Publications

• S. N. Srirama, A. Ostovar: Optimal Resource Provisioning for Scaling Enterprise Applications on the Cloud, The 6th IEEE International Conference on Cloud Computing Technology and Science (CloudCom-2014), December 15-18, 2014, pp. 262-271. IEEE.
• S. N. Srirama, J. Viil: Migrating Scientific Workflows to the Cloud: Through Graph-partitioning, Scheduling and Peer-to-Peer Data Sharing, 16th IEEE International Conference on High Performance and Communications (HPCC 2014) workshops, August 20-22, 2014, pp. 1137-1144. IEEE.

Master theses

Alireza Ostovar – Optimal Resource Provisioning for Workflows in Cloud (2014)

Bachelor theses

Jaagup Viil – Remodelling Scientific Workflows for Cloud (2014)

A fault tolerant BSP framework on Hadoop YARN

We looked for a distributed computing framework that could be utilized for running iterative scientific computing algorithms in the cloud, but most solutions proved lacking in some respect. This is why we decided to create our own with the following goals:

• Provide automatic fault recovery.
• Retain the program state after fault recovery.
• Provide a convenient programming interface.
• Support (iterative) scientific computing applications.

The following is an example of a linear system of equations solver, using the conjugate gradient method, implemented in NEWT:

• One has to register the program definition with the application master.

public class CGTestMaster {
	public static void main(String[] args) {
		//simply initialize the default configuration
		Configuration conf = NEWTConfiguration.createDefault();
		//the test data will be generated by the worker nodes
		//so only use commandline arguments as input and 
		//specify the number of processes to request
		Input input = new CommandLineInput(conf, args, Integer.parseInt(args[0]));
		//create an application master, using the configuration,
		//the state class definition and the input
		FaultTolerantJobMaster am =
				new FaultTolerantJobMaster(conf, CGState.class, input);
		//register the application code with the master
		am.addStage("init", new CG.Init());
		am.addStage("beginLoop", new CG.BeginLoop());
		am.addStage("checkCondition", new CG.CheckCondition());
		am.addStage("doMatVec", new CG.DoMatVec());
		//register our custom message type
		am.registerMessageType(DoubleArrayWritable.class);
		//finally, run the application master
		am.run();
	}
}

The CGState.class contains the definitions of all state variables used by the algorithm and instructions to serialize them to a bitstream. The objects that are given as arguments to am.addStage contain the code of the program, an example of these stages is:

//verbose Java 1.6 "closure" definition
//can be made much more concise with Java 1.7
//or a Scala interface
public static class CheckCondition extends Stage {
	//the method takes the BSPComm communicatior,
	//the state object and the superstep number as arguments
	@Override
	public String execute(BSPComm bsp, CGState state, int superstep) {
		//we know that on the previous superstep each process
		//sent a message of 2 doubles, containing the partial
		//dot product and local maximum norm of the residual
		//vector (which is used as an error estimate)
		double[] recv = getFromAll(bsp, 2);
		state.u += recv[0];
		state.error = Math.max(state.error, recv[1]);
		//depending on some state variables, determine what should be done next
		if (state.error > TOLERANCE && state.it <= MAX_IT) {
			if (state.it == 0) {
				for (int i = 0; i < state.p.size(); i++) {
					state.p.getData()[i] = state.z.getData()[i];
				}
			} else {
				double v = state.u/state.ou;
				CGMath.scalevector(state.p.getData(), v);
				CGMath.addVector(state.p.getData(), state.z.getData());
			}
			//we need to synchronize the vector p for matrix vector multiplication
			sendToNeighbors(bsp, state, state.p.getData());
			//do matrix vector multiplication on the next superstep
			return "doMatVec";
		} else {
			//we're done, write result to disk and stop
			return Stage.STAGE_END;
		}
	}
}

Here the getFromAll and sendToNeighbors are implemented using the communicator’s send function to transmit parts of some of the state from the CGState object. Since these are common communication patterns, these might eventually be implemented on the framework level. Some mathematical operations that need to be performed on the state variables are implemented in a separate class CGMath.

• To run the program, it has to be submitted to the YARN cluster:

public class TestJob {
	public static void main(String[] args) {
		NEWTClient client = new NEWTClient();
		client.submit(CGTestMaster.class, args, true);
		client.close();
	}
}

For some general information on NEWT, refer to this poster.

While the framework is not in the shape where it can be used for any serious projects yet, it’s open-source and available at: https://bitbucket.org/mobilecloudlab/newt

Aneka plays the role of Application Platform as a Service for Cloud Computing. Aneka supports various programming models involving Task Programming, Thread Programming and MapReduce Programming and tools for rapid creation of applications and their seamless deployment on private or public Clouds to distribute applications. The goal of this study is to use the Aneka platform to establish hybrid clouds for solving scientific computing problems. We have set up a local Aneka Cloud in the University of Tartu and the idea is to also involve resources from public clouds such as Amazon EC2 when larger scientific computing experiments need to be executed.

People

• Assoc. Prof. Dr. Satish Srirama
• Pelle Jakovits
• Dr. Suraj Pandey
  • Current: Research Scientist, IBM Australia Research Lab @ 204 Lygon Street, Melbourne, Australia
  • Previously: The Cloud Computing and Distributed Systems (CLOUDS) Laboratory, University of Melbourne
• Dr. Rajkumar Buyya
  • Professor of Computer Science and Software Engineering, University of Melbourne

REuse and Migration of legacy applications to Interoperable Cloud Services (REMICS).

REMICS is an EU FP7 project. The goal of REMICS is to develop advanced model-driven methodology and tools for REuse and Migration of legacy applications to Interoperable Cloud Services. Combining the deployment models of the cloud paradigm with the SaaS (Software as a Service) model allows software companies to reuse and extend their legacy applications, compose them in new ways, and reach new markets. REMICS also takes an active role in the standardization of metamodels and languages in the OMG activities around Architecture-Driven Modernization, SOA and Cloud Computing, Model-Driven Interoperability, Models@Runtime, Model Checking and Model-based Testing. Within the project, we are dealing with migrating legacy OLAP/OLTP applications to the cloud frameworks. For this purpose, we are developing a desktop to cloud migration (D2CM) tool. The tool integrates several software libraries to facilitate virtual machine image migration and management of experiments running in the cloud. We are also studying and developing a framework for verifying the performance and scalability of applications running in the cloud.

Homepage

People

• Satish Srirama
• Michele Mazzucco
• Pelle Jakovits
• Martti Vasar
• Huber Flores

Publications

• M. Vasar, S. N. Srirama, M. Dumas: Framework for Monitoring and Testing Web Application Scalability on the Cloud, Nordic Symposium on Cloud Computing & Internet Technologies (NORDICLOUD 2012), Co-located event at Joint 10th Working IEEE/IFIP Conference on Software Architecture & 6th European Conference on Software Architecture (WICSA/ECSA 2012), August 20-24, 2012. (Accepted for publication)
• M. Mazzucco, M. Vassar, M. Dumas: Squeezing out the Cloud via Profit-Maximizing Resource Allocation Policies, In Proceedings of the 20th IEEE International Symposium on Modeling, Analysis and Simulation of Computer and Telecommunication Systems (MASCOTS), Arlington, VA, USA, August 2012. IEEE Computer Society.
• M. Mazzucco, I. Mitrani: Empirical Evaluation of Power Saving Policies for Data Centers, ACM Greenmetrics, London (UK), June 2012.

Theses

• M. Vasar, A Framework for Verifying Scalability and Performance of Cloud Based Web Applications, Master’s thesis, University of Tartu, June, 2012. Supervisor: Satish Srirama

Technological achievements in user context monitoring techniques enable automating certain computational tasks which are executed by anticipating the user’s intention. These kind of techniques are based on the processing of sensor information that is collected from multiple sources such as smartphones, environment, etc.

In the case of smartphones, sensor information is gathered by embedded micromechanical artifacts (e.g. accelerometer, gyroscope, etc) and processed locally in real-time (in the phone) for changing some usability aspects in the mobile applications. For instance, the accelerometer sensor can be used for rotating the screen of the device depending on how the user is holding the handset. Another example is the light sensor which enables augmenting and decreasing the brightness of the mobile screen according to the situation of the environment (e.g. indoor, outdoor, etc).

Similarly, environmental sensor information is provisioned as a service (raw data) to mobile users by locating multiple microelectromechanical appliances within the environment. For instance, a thermistor sensor can be located for perceiving the temperature in the context of the user so that a mobile application can use that information for a proactive reaction (e.g. displaying different background screens or triggering a vibration event in the case of mobile pervasive games).

Research staff

• Assoc. Prof. Dr. Satish Narayana Srirama
• Huber Flores
• Carlos Paniagua

Projects

• Tanel Tähepõld – Context-aware Mobile Games using Android, Arduino and HTML5 (2012)
• Martti Marran – Generating Schematic Indoor Representation based on Context Monitoring Analysis of Mobile Users (2012)

Scientific computing applications usually need huge amounts of computational power. The cloud provides interesting high-performance computing solutions, with its promise of virtually infinite resources.  However, migrating scientific computing problems to clouds and the re-creation of a software environment on the vendor-supplied OS and cloud instances is often a laborious task. It is also assumed that the scientist who is performing the experiments has significant knowledge of computer science, cloud computing and the migration procedure. Most often, this is not the case. Considering this obstacle, we have designed tools that help scientists to migrate their applications to the cloud. The idea is to migrate the complete desktop environment, with which the scientist is working daily, to the cloud directly. The developed desktop-to-cloud-migration (D2CM) tool supports transformation of virtual machine images, deployment description and life-cycle management for applications to be hosted on Amazon’s Elastic Cloud Computing (EC2) or compatible infrastructure such as Eucalyptus. We used an electrochemical case study which extensively used the tool in drawing domain specific results. From the analysis, it was observed that D2CM tool not only helps in the migration process but also helps in optimizing the experiments, clusters and thus the costs for conducting scientific computing experiments on the cloud.

This work is done in the context of the EU FP7 project REMICS – “Reuse and Migration of legacy applications to Interoperable Cloud Services”

Software and guides

Source code is available at: https://bitbucket.org/mobilecloudlab/d2cm

User guide: http://mc.cs.ut.ee/mcsite/projects/d2cm-user-guide-v1.0/at_download/file

People

• Satish Srirama
• Chris Willmore
• Pelle Jakovits
• David Rodas
• Vladislav Ivaništšev, Center for Interface Science and Catalysis, University of Strathclyde, UK.
• Enn Lust, Faculty of Science and Technology, Institute of Chemistry, University of Tartu.

Publications

• S. N. Srirama, V. Ivanistsev, P. Jakovits, C. Willmore: Direct Migration of Scientific Computing Experiments to the Cloud, The 2013 (11th) International Conference on High Performance Computing & Simulation (HPCS 2013), July 01-05, 2013, pp. 27-34. IEEE. (Nominee for the Outstanding Paper Award)
• S. N. Srirama, C. Willmore, V. Ivanistsev, P. Jakovits et al.: Desktop to Cloud Migration of Scientific Experiments, 2nd International Workshop on Cloud Computing and Scientific Applications (CCSA) @ 12th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid 2012), May 13-16, 2012. (* No proceedings) Slides

The goal of the Stratus framework is to provide a distributed computing platform and tools for performing large scale scientific computing simulations and experiments. The main reason for designing a new cloud computing framework aimed for scientific simulations is that the existing frameworks that we have studied so far (Hadoop, Spark, Twister, etc.), do not provide full support for iterative distributed applications and especially iterative scientific simulations. Additionally, distributed computing frameworks are usually designed for static computer clusters and do not take into account characteristics that have made cloud computing a successful source for computing infrastructure.We find that exploiting cloud computing characteristics like elasticity, agility, scalability and the ability to provision finegrained computing resource on demand and nearly on realtime can greatly improve the applicability of such frameworks and would lower the costs of deploying applications in the public cloud.

People

• Satish Srirama
• Pelle Jakovits
• Ilja Kromonov

Publications

• P. Jakovits, S. N. Srirama, I. Kromonov: Stratus: A distributed computing framework for scientific simulations on the cloud, The Fifth International Symposium on Advances Of High Performance Computing And Networking (AHPCN-2012), In conjunction with The 14th IEEE International Conference on High Performance Computing and Communications (HPCC-2012), 25-27 June, 2012, pp. 1053-1059. IEEE.
• P. Jakovits, S. N. Srirama: Stratus: Scientific simulations in the cloud, The 12th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing, May 13-16, 2012.

Posters

• Stratus: scientific simulations in the cloud
  

In this project a home care robot is built, which can approach and initiate communication with the patient being monitored, thus validating the threat of the alerts detected by the sensors attached to the patient. Thus, this robot allows monitoring risk patients such as elderly people, without the need of a caregiver, and allowing them the opportunity to stay at home independently. Scenarios such as the fall detection of elderly patients were demonstrated as part of the project.

Homepage

A youtube video of the first public demonstration of the ROBO M.D. system by Dr. Satish Srirama

“Excuse Me, Have You Fallen Down?”, a blog at UT Blog

Presentations

• S. N. Srirama: ROBO M.D. – Home Care Robot for monitoring and detection of critical situations, New Ways to Competitiveness conference 2010, Tartu, Estonia, May 10-12, 2010.
• Slides with final results

Partners

1. JKU (UAT): Johannes Kepler University, Institute for Design and Control of Mechatronical Systems
2. IEIIT (LOM): Italian National Research Council (CNR) Institute of Electronic, Information and  Communication Technologies (IEIIT) Milano Branch(MI)
3. Fontys (NBR): Fontys University of Applied Sciences
4. JU (SWB): University of South Bohemia in České Budějovice
5. UT (Tartu): University of Tartu

Mobile cloud applications are considered as the next generation of mobile applications, due to their promises concerning the efficient utilization of mobile resources (e.g. offloading when it is needed), the rich functionality that can scale on demand and the reasonable levels of real-time interactivity for managing data-intensive tasks. However, adapting the cloud paradigm for mobile devices is still in its infancy and several issues are to be answered first. Some of the prominent questions are; how to decrease the effort and complexity of developing a mobile application that requires accessing distributed hybrid cloud architectures? How to handle a multi-cloud operation without overloading the mobile resources? How to keep the properties (e.g. memory use, application size etc.) of a mobile cloud application similar to that of a native one?

MCM and the resource intensive tasks can easily be envisioned in several scenarios. For instance, we have developed several mobile applications that benefit for going cloud-aware. Zompopo, consists of the provisioning of context-aware services for processing data collected by the accelerometer with the purpose of creating an intelligent calendar. CroudSTag, consists of the formation of a social group by recognizing people in a set of pictures stored in the cloud. Finally, Bakabs is an application that helps in managing the cloud resources themselves from the mobile, by applying linear programming techniques for determining optimized cluster configurations for the provisioning of Web-based applications.

Research staff

Assoc. Prof. Dr. Satish Narayana Srirama

Huber Flores

Publications

• H. Flores, S. N. Srirama, C. Paniagua: Towards Mobile Cloud Applications: Offloading Resource-Intensive Tasks To Hybrid Clouds, International Journal of Pervasive Computing and Communications, ISSN: 1742-7371. Emerald Group Publishing Limited. (In Print)
• S. N. Srirama, C. Paniagua, H. Flores: Social Group Formation with Mobile Cloud Services, Service Oriented Computing and Applications Journal, ISSN: 1863-2386. Springer. DOI: 10.1007/s11761-012-0111-5 (In Print)
• C. Paniagua, S. N. Srirama, H. Flores: Bakabs: Managing Load of Cloud-based Web Applications from Mobiles, The 13th International Conference on Information Integration and Web-based Applications & Services (iiWAS-2011), December 5-7, 2011, pp. 489-495. ACM.

• S. N. Srirama, C. Paniagua, H. Flores: CroudSTag: Social Group Formation with Facial Recognition and Mobile Cloud Services, The 8th International Conference on Mobile Web Information Systems (MobiWIS 2011), September 19-21, 2011, v. 5 of Procedia Computer Science, pp. 633-640. Elsevier. doi: 10.1016/j.procs.2011.07.082.
• S. N. Srirama, H. Flores, C. Paniagua: Zompopo: Mobile Calendar Prediction based on Human Activities Recognition using the Accelerometer and Cloud Services, 5th International Conference on Next Generation Mobile Applications, Services and Technologies (NGMAST 2011), September 14-16, 2011, pp. 63-69. IEEE.

Master theses

Oleg Petshjonkin – Migration of Native Android Applications to HTML5 (2012)

Former staff

Carlos Paniagua

Hybrid cloud and cloud interoperability are essential for mobile scenarios in order to foster the decoupling of the handset to a specific cloud vendor, to enrich the mobile application with the variety of cloud services provided on the Web and to create new business opportunities and alliances. However, developing a mobile cloud application involves adapting different Web APIs from different cloud vendors within a native mobile platform. To counter the problems with interoperability across multiple clouds, to perform data-intensive processing invocation from the handset and to introduce the platform independence feature for the mobile cloud applications, we have designed a Mobile Cloud Middleware (MCM).

MCM abstracts the Web API of different cloud levels (multiple vendors) using Clojure and provides a unique interface that responds (JSON-based) according to the cloud services requested (REST-based).

Research staff

Assoc. Prof. Dr. Satish Narayana Srirama

Kristjan Reinloo

Publications

H. Flores, S. N. Srirama, C. Paniagua: A Generic Middleware Framework for Handling Process Intensive Hybrid Cloud Services from Mobiles, The 9th International Conference on Advances in Mobile Computing & Multimedia (MoMM-2011), December 5-7, 2011, pp. 87-95. ACM.

Master theses

Huber Flores – Mobile Cloud Middleware (2011)

Former staff

Carlos Paniagua
Huber Flores