Vladimir Vlassov

Peer-to-Peer resource discovery in Grids: Models and systems

Resource location or discovery is a key issue for Grid systems in which applications are composed of hardware and software resources that need to be located. Classical approaches to Grid resource location are either centralized or hierarchical and will prove inefficient as the scale of Grid systems rapidly increases. On the other hand, the Peer-to-Peer (P2P) paradigm emerged as a successful model that achieves scalability in distributed systems. One possibility would be to borrow existing methods from the P2P paradigm and to adopt them to Grid systems taking into consideration the existing differences. Several such attempts have been made during the last couple of years. This paper aims to serve as a review of the most promising Grid systems that use P2P techniques to facilitate resource discovery in order to perform a qualitative comparison of the existing approaches and to draw conclusions about their advantages and weaknesses. Future research directions are also discussed.

MapReduce: Limitations, Optimizations and Open Issues

MapReduce has recently gained great popularity as a programming model for processing and analyzing massive data sets and is extensively used by academia and industry. Several implementations of the MapReduce model have emerged, the Apache Hadoop framework being the most widely adopted. Hadoop offers various utilities, such as a distributed file system, job scheduling and resource management capabilities and a Java API for writing applications. Hadoops success has intrigued research interest and has led to various modifications and extensions to the framework. Implemented optimizations include performance improvements, programming model extensions, tuning automation and usability enhancements. In this paper, we discuss the current state of the Hadoop framework and its identified limitations. We present, compare and classify Hadoop/MapReduce variations, identify trends, open issues and possible future directions.

ElastMan: elasticity manager for elastic key-value stores in the cloud

The increasing spread of elastic Cloud services, together with the pay-as-you-go pricing model of Cloud computing, has led to the need of an elasticity controller. The controller automatically resizes an elastic service in response to changes in workload, in order to meet Service Level Objectives (SLOs) at a reduced cost. However, variable performance of Cloud Virtual Machines and nonlinearities in Cloud services, such as the diminishing reward of adding a service instance with increasing the scale, complicates the controller design. We present the design and evaluation of ElastMan, an elasticity controller for Cloud-based elastic key-value stores. ElastMan combines feedforward and feedback control. Feedforward control is used to respond to spikes in the workload by quickly resizing the service to meet SLOs at a minimal cost. Feedback control is used to correct modeling errors and to handle diurnal workload. To address nonlinearities, our design of ElastMan leverages the near-linear scalability of elastic Cloud services in order to build a scale-independent model of the service. We have implemented and evaluated ElastMan using the Voldemort key-value store running in an OpenStack Cloud environment. Our evaluation shows the feasibility and effectiveness of our approach to automation of Cloud service elasticity.

Router Placement in Wireless Sensor Networks

In this paper we propose and evaluate algorithms for placement of routers in a wireless sensor network. There are two major requirements on router placement. First, a placement must guarantee connectivity, i.e. every sensor must be able to communicate through routers with a predefined computer-connected gateway node. Second, a placement must provide robust communication in the case of router failures. This is achieved by placing redundant routers that increase the number of possible routes. Both requirements should be met by placing as few routers as possible. The proposed algorithms compute placement in an efficient and reasonably fast way

Supporting cloud deployment in the Guifi.net community network

Community networking is an emerging model of a shared communication infrastructure in which communities of citizens build and own open networks. Community networks offer successfully IP-based networking to the user. Cloud computing infrastructures however, while common in todays Internet, hardy exist in community networks. We explain our approach to bring clouds into the Guifi.net community network. For this we have started integrating part of our cloud prototype into the Guifi.net community network management tools. A proof-of-concept cloud infrastructure is currently under deployment in the Guifi.net community network. Our long term vision is that the users of community networks will not need to consume cloud applications from the Internet, but find them within the community network.

Stay-Away , protecting sensitive applications from performance interference

While co-locating virtual machines improves utilization in resource shared environments, the resulting performance interference between VMs is difficult to model or predict. QoS sensitive applications can suffer from resource co-location with other less short-term resource sensitive or batch applications. The common practice of overprovisioning resources helps to avoid performance interference and guarantee QoS but leads to low machine utilization. Recent work that relies on static approaches suffer from practical limitations due to assumptions such as a priori knowledge of application behaviour and workload. To address these limitations, we present Stay-Away, a generic and adaptive mechanism to mitigate the detrimental effects of performance interference on sensitive applications when co-located with batch applications. Our mechanism complements the allocation decisions of resource schedulers by continuously learning the favourable and unfavourable states of co-execution and mapping them to a state-space representation. Trajectories in this representation are used to predict and prevent any transition towards interference of sensitive applications by proactively throttling the execution of batch applications. The representation also doubles as a template to prevent violations in the future execution of the repeatable sensitive application when co-located with other batch applications. Experimental results with realistic applications show that it is possible to guarantee a high level of QoS for latency sensitive applications while also improving machine utilization.

A Design Methodology for Self-Management in Distributed Environments

Autonomic computing is a paradigm that aims at reducing administrative overhead by providing autonomic managers to make applications self-managing. In order to better deal with dynamic environments, for improved performance and scalability, we advocate for distribution of management functions among several cooperative managers that coordinate their activities in order to achieve management objectives. We present a methodology for designing the management part of a distributed self-managing application in a distributed manner. We define design steps, that includes partitioning of management functions and orchestration of multiple autonomic managers. We illustrate the proposed design methodology by applying it to design and development of a distributed storage service as a case study. The storage service prototype has been developed using the distributing component management system Niche. Distribution of autonomic managers allows distributing the management overhead and increased management performance due to concurrency and better locality.

Performance Characterization of In-Memory Data Analytics on a Modern Cloud Server

In last decade, data analytics have rapidly progressed from traditional disk-based processing to modern in-memory processing. However, little effort has been devoted at enhancing performance at micro-architecture level. This paper characterizes the performance of in-memory data analytics using Apache Spark framework. We use a single node NUMA machine and identify the bottlenecks hampering the scalability of workloads. We also quantify the inefficiencies at micro-architecture level for various data analysis workloads. Through empirical evaluation, we show that spark workloads do not scale linearly beyond twelve threads, due to work time inflation and thread level load imbalance. Further, at the micro-architecture level, we observe memory bound latency to be the major cause of work time inflation.

Parallel Agent-Based Simulation on a Cluster of Workstations

We discuss a parallel implementation of an agent-based simulation. Our approach allows to adapt a sequential simulator for large-scale simulation on a cluster of workstations. We target discrete-time simulation models that capture the behavior of WWW. The real-world phenomena of emerged aggregated behavior of the Internet population is studied. The system distributes data among workstations, which allows large-scale simulations infeasible on a stand-alone computer. The model properties cause traffic between workstations proportional to partition sizes. Network latency is hidden by concurrent simulation of multiple users. The system is implemented in Mozart that provides multithreading, dataflow variables, component-based software development, and network-transparency. Currently we can simulate up to 106 Web users on 104 Web sites using a cluster of 16 computers, which takes few seconds per simulation step, and for a problem of the same size, parallel simulation offers speedups between 11 and 14.

ElastMan: autonomic elasticity manager for cloud-based key-value stores

The increasing spread of elastic Cloud services, together with the pay-as-you-go pricing model of Cloud computing, has led to the need of an elasticity controller. The controller automatically resizes an elastic service in response to changes in workload, in order to meet Service Level Objectives (SLOs) at a reduced cost. However, variable performance of Cloud virtual machines and nonlinearities in Cloud services complicates the controller design. We present the design and evaluation of ElastMan, an elasticity controller for Cloud-based elastic key-value stores. ElastMan combines feedforward and feedback control. Feedforward control is used to respond to spikes in the workload by quickly resizing the service to meet SLOs at a minimal cost. Feedback control is used to correct modeling errors and to handle diurnal workload. We have implemented and evaluated ElastMan using the Voldemort key-value store running in a Cloud environment based on OpenStack. Our evaluation shows the feasibility and effectiveness of our approach to automation of Cloud service elasticity.