This page summarizes the research results of Cloud computing group in the MTA-SZTAKI LPDS. First, we introduce the researchers who are participating in the research group. Then, we summarize the goals, research interests and future works of the group. Afterwards, we shortly revisit our past and present project participation. Finally, we proceed with the discussion of our research results organized around 5 main topics: (i) Desktop grid virtualization, (ii) Virtual appliance delivery, (iii) Cloud federations, (iv) Cluster scaling and (v) Clouds and nature.

People

The research group was founded in March 2011. Based on their previous research results and interests the following members joined the group so far:

• Sandor Acs - Cluster scaling
• Gabor Kecskemeti - Current group coordinator, intersts in cloud federations and appliance delivery
• Attila Kertesz - Cloud federations
• Miklos Kozlovszky
• Attila Marosi - Cloud federations, cluster scaling and desktop grid virtualization
• Zsolt Nemeth - Clouds and nature
• Zoltan Farkas

Research interests

Our cloud computing group is focusing on research related to the field of Infrastructure as a Service (IaaS) based cloud systems. We have investigated various applications of virtualization techniques (including several virtual machine monitors - e.g. Xen, Qemu, Kvm). We have provided techniques to extend the currently available desktop grid systems with behavior similar to IaaS systems in order to enable the creation of virtual machines on the computing resource donor machines (see details here). We have faced the problem of distributing virtual machine images (or virtual appliances) between the different components of the resulting system. This problem has also arisen while we were focusing on virtualization based service deployment solutions. Our research revealed that specially crafted and stored virtual appliances could improve their delivery significantly (see details here). Efficient and automated virtual appliance deployment provides the foundation to federated cloud infrastructures and auto-scaling clusters. Therefore, we have analyzed techniques to enable interoperation between the currently available IaaS systems through multi level brokering solutions (see details here). We have also studied approaches that allow localized computing cluster infrastructures (accessible through batch systems) to be scaled by creating special virtual appliances supporting the extension towards multiple IaaS systems (see details here). Our research results identified several autonomous, heuristic and optimization problems that we plan to investigate with nature-inspired models and algorithms (see details here).

Research projects

The group members participate in several on going projects of the laboratory. The first project where our cloud expertise was utilized is the S-Cube project. It started in 2008, and the group members are mainly focusing on how to increase SLA awareness of cloud infrastructures. Next, the group also participates in the research and development efforts of the EDGI project, where our members provide a support role in the context of bridging (service and desktop) grid systems and infrastructure clouds. Finally, the group also provides support for the cloud related aspects of the recently started SCI-BUS project.

Research results

Virtualization in Desktop Grid Systems

The aim of Desktop Grid Systems (DGS's), especially volunteer computing, is to harvest the idle cycles of home (non-dedicated) computers. These computers differ in many ways (e.g.: CPU architecture, memory capacity and operating system) and some scientific applications are either a.) legacy (no source code available) or b.) too complex to port to a specific Distributed Computing Infrastructure (DCI) like DGS's. Virtualization aims to solve these problems by providing a homogeneous environment, which hides the specifics of the DCI and the host resource (CPU, operating system) from the application. This environment also allows using specific VA (Virtual Appliances) for each application, thus providing a customized execution environment. Our goal is to extend the range of scientific applications for DGS's and to minimize the effort required for porting to these DCI's [pdp10].

Virtual appliance delivery optimization

The use of virtual appliances could provide a flexible solution to services deployment if their slow deployment time and inefficient creation methods could be avoided. We have tackle the problem of deployment time by virtual appliance distribution technique that first identifies appliance parts and their internal dependencies. Then based on service demand our technique efficiently distributes the identified parts to virtual appliance repositories. The inefficiencies of the virtual appliance creation process are targeted with the Automated Virtual appliance creation Service (AVS - [fgcs]) that can extract and publish an already deployed service for the developer. This recently acquired virtual appliance is optimized for service deployment time with the virtual appliance optimization facility [vaopt] that removes the non-functional parts of the appliance.

Management of cloud federations

Building on our former research results [mb], one of our goals is to manage autonomously a federation of heterogeneous distributed systems [pdp11a] with advanced deployment and brokering techniques. As Cloud Computing infrastructure solutions are becoming increasingly popular, we revise our solutions to meet the emerging demands of cloud-based environments. In our first research results, we revealed the effective virtual machine destination selection and management capabilities of our Federated Cloud Management architecture in [ccr11]. This architecture is built on efficient meta-brokering, cloud brokering, cloud resource management and automatic service deployment. The decision making process in the various architectural components is supported with an integrated cloud montioring service as discussed in [pdp12b]. We also investigated the scenarios that require the autonomous behavior of cloud federations in [pdp12a]. We plan to provide a uniform, transparent interface for users to a federation across multiple cloud providers, including both public and private clouds.

Auto-scaling clusters in the cloud

IaaS systems provide a flexible way to either extend existing or create on-demand pool of resources (e.g.: clusters). We investigated the possible methods for this [pdp11b] and developed a system that is able to utilize the de-facto standard Amazon EC2 interface (supported by OpenNebula, Amazon EC2, Eucalyptus and OpenStack) to manage resources for virtual clusters pooled by a batch system (e.g.: Condor, BOINC or PBS) and to support automatic service deployments. As a result, we can use the advantages of these batch systems with or without a locally available cluster. For example, our techniques are capable to extend an already deployed PBS system and support the seamless integration of legacy PBS applications with cloud systems. We currently focus on investigating different resource life-cycle-management scenarios and intra-cloud brokering. Another goal of ours is to provide SLA's for pools of volatile resources by allocating on-demand dedicated resources. Finally, we have also shown in [cloudcom2011] that automatic cluster scaling could significantly reduce the makespan of scientific workflows executed on desktop grids and on IaaS systems.

Clouds and nature

As the size and complexity of distributed systems is growing, there is an inherent need for entities that are able to recognise various situations, control their behaviour, potentially adapt to changing conditions in such a way that its working parameters are within predefined boundaries and is able to fulfil its goals. Such entities are often referred as autonomic systems and are characterised by some of the self-aware, self-coordinating, self-adapting, self-healing, self-protecting, self-configuring and other properties.

A new research direction is aimed at investigating non-conventional approaches to tackle with coordination, control, optimisation, adaptation and other arising issues in complex, large-scale and heterogeneous systems, tightly coupled to questions of autonomic systems. In our initial work, we considered a chemical modelling approach and studied its features in a (grid) workflow [sefm] and service composition problems [soarch, europar11]. In the future we plan to analyse scenarios that involve autonomic aspects in cloud computing and investigate the applicability of nature-inspired models and algorithms to them.