There has been tremendous growth in the amount of visual data available on the Internet in recent years. One type of visual data of particular interest is produced by network cameras providing real-time views. Millions of network cameras around the world continuously stream data to viewers connected to the Internet. This data may be used by a wide variety of applications such as enhancing public safety, urban planning, emergency response, and traffic management which are computationally intensive. Analyzing this data requires significant amounts of computational resources. Cloud computing can be a preferred solution for meeting the resource requirements for analyzing these data. There are many options when selecting cloud instances (amounts of memory, number of cores, locations, etc.). Inefficient provisioning of cloud resources may become costly in pay-per-use cloud computing. This paper presents a method to select cloud instances in order to meet the performance requirements for visual data analysis at a lower cost. We measure the frame rates when analyzing the data using different computer vision methods and model the relationships between frame rates and resource utilizations. We formulate the problem of managing cloud resources as a Variable Size Bin Packing Problem and use a heuristic solution. Experiments using Amazon EC2 validate the model and demonstrate that the proposed solution can reduce the cost up to 62 percent while meeting the performance requirements.

High performance computing systems comprised of hundreds or thousands of computational nodes can generate a high volume of system log entries at a high data velocity. Analyzing these logs soon after they are generated is a significant challenge, due to the complexity of log messages, the speed at which they are produced, and the lack of a method to quickly map or categorize messages to meaningful sets. The impact of this problem is that it is not possible to comprehensively glean timely information from logs about the overall system or the health of individual nodes. In this paper, we address this problem through the development of a novel approach for system log analysis based on a markov random field (MRF) that can quickly categorize system log messages into multiple categories based on representative training examples provided by a user. We present a theoretical model of our approach, followed by an extensive evaluation of the accuracy and performance of the implementation of our model. We found that our MRF based approach can quickly categorize system log messages with a high degree of accuracy.

Indoor localization can provide a number of different services such as location-aware advertisement, indoor navigation and automating different appliances based on the user location. A number of different techniques such as time-difference- of-arrival, angle-of-arrival, time-of-flight, and received signal strength indicator (RSSI) have been used to provide Location Based Services (LBS). RSSI is one of the widely used methods as it is cost efficient and easy to implement. However, RSSI's performance is limited by multipath fading and indoor noise. Particle Filter (PF) is an accurate Bayesian Filtering algorithm that can improve the performance of RSSI-based indoor localization. However, PF is not able to satisfy the high accuracy requirement (possibly 10cm) of indoor localization. In this paper, we present Particle Filter-Extended Kalman Filter (PFEKF) cascaded algorithm that combines PF and EKF in series to reduce the impact of multipath effects and noise on the RSSI. Our experimental results show that PFEKF improves the localization accuracy by 31.3% and 33.9% in 3D and 2D environments respectively when compared with using only a PF.

Modern web/mobile applications have a highly variable user load from different geographical locations that depends upon the time-of-day, flash crowds, and growth rate (high or low) of user load. Irrespective of server hosting location of the applications, users from different geographical regions need highly responsive applications because bad performance results in unhappy customers. A recent survey suggests that mobile applications lose 34% of their users due to poor response time. To enhance the user experience of applications, we propose Global Live Migration (GLM) of VMs, which allows VMs to be migrated between different geographical regions over Wide Area Networks (WANs). Our experimental setup consists of Software-defined Network (SDN) enabled networking infrastructure because of its extensive control over network traffic routing. Our objective is to place the VMs on sites closest to the region with a major population of respective application users. However, there are high chances of congestion in inter-site links if a random migration sequence is followed while performing multiple VM migrations that may result in migration failures. We formulate a Mixed Integer Programming (MIP), and solve it using a Lagrangian relaxation based subgradient technique. We also propose a feedback-based control system with self-tuning regulators to improve the migration sequence regularly. Our extensive simulations show that our proposed solution achieves a minimum improvement of 38.09% compared to the Best Fit algorithm.

Proximity-based Services (PBS) require high detection accuracy, energy efficiency, wide reception range, low cost and availability. However, most existing technologies cannot satisfy all these requirements. Apple's Bluetooth Low Energy (BLE), named iBeacon, has emerged as a leading candidate in this domain and has become an almost industry standard for PBS. However, it has several limitations. It suffers from poor proximity detection accuracy due to its reliance on Received Signal Strength Indicator (RSSI). To improve proximity detection accuracy of iBeacons, we present two algorithms that address the inherent flaws in iBeacon's current proximity detection approach. Our first algorithm, Server-side Running Average (SRA), uses the path-loss model-based estimated distance for proximity classification. Our second algorithm, Server-side Kalman Filter (SKF), uses a Kalman filter in conjunction with SRA. Our experimental results show that SRA and SKF perform better than the current moving average approach utilized by iBeacons. SRA results in about a 29% improvement while SKF results in about a 32% improvement over the current approach in proximity detection accuracy.

With the increasing growth in the amount of information stored on remote locations and cloud systems, many service providers are seeking ways to reduce the amount of redundant information. Data deduplication can reduce the network traffic without loss of information, and consequently increase the available network bandwidth. However, due to the heavy computation overhead for detecting and reducing the redundant data, deduplication itself can become a bottleneck in high capacity links. In this paper, we propose a method named Hardware Accelerated Redundancy Elimination in Network Systems (HARENS). HARENS can significantly improve the performance of redundancy elimination in a network system by leveraging General Purpose Graphic Processing Unit (GPGPU) optimizations, as well as other optimizations such as the use of a hierarchical multi-threaded pipeline, Hash-Match, and memory efficiency techniques. Our results indicate that throughput can be increased by a factor of 14 compared to a native implementation of a network deduplication algorithm, providing a net transmission increase of up to 10.7 Gigabits per second (Gbps).

The use of video data for a variety of applications has gained immense popularity. These applications include traffic monitoring, surveillance, retail store management, etc. Thousands of publicly accessible network cameras distributed around the world are potential sources of video data. The applications analyzing data from network cameras have different resource requirements (CPU, memory, etc.) and performance requirements (video frame rate). A preferred way to meet these requirements is to use a cloud infrastructure and a pay-per-use model of the cloud. Cloud computing offers resources, referred to as cloud instances, with different capacities and at different locations. The cost of these instances are dependent on their capacities and locations. The frame rate of the video data obtained from a network camera impacts the accuracy of the analysis and is dependent on the location of the cloud instance. Hence it is important to select the locations of the instances to meet the application performance requirements. This paper presents a resource management approach to select the locations, types, and number of cloud instances to analyze real-time video data from the network cameras while meeting the performance requirements and reducing the analysis cost. We model the resource management problem as a variable size bin packing problem and describe a heuristic algorithm to find a solution. This paper uses Amazon EC2 to evaluate our new resource manager, and observes that our method can reduce the analysis cost up to 56% compared with two other strategies for selecting instance locations.

In the competitive world of modern web applications, performance plays a crucial role. An e-commerce company estimated that every 100ms delay reduces sales by 1 percent, and a popular search engine reported that every 500ms delay in search reduces earnings by 20 percent. The demands from users for these services can vary widely based on factors such as the time-of-day and unexpected events that can trigger flash crowds. To meet these demands web applications can be organized using a multi-tier architecture to make them modular and scalable in a cloud environment. However, a highly dynamic workload and different types of resource requirements in each tier can make it difficult to model the behavior of these applications. This presents two significant challenges to infrastructure providers: 1) to model the behavior of an application workload and provide responsive resources using dynamic resource provisioning, and 2) to maintain performance-based (response time) Service Level Agreements (SLAs). In this paper, we formulate a convex optimization problem for resource allocation, and offer a strict SLA for performance. We adopt an SLA violation cost model to formulate our optimization problem and derive the solution for dynamic resource provisioning. To achieve a strict SLA for the response time of an application, we propose a predictive model that seeks to dynamically provision resources using a Feedback-based Control System (FCS). Our model is applicable for a broad range of multi-tier applications. We demonstrate the effective use of our model through experiments that analyze the behavior of an online auction application using a common workload benchmark.

The visual data generated by network cameras can be valuable for a wide range of scientific studies such as weather, wildlife, and traffic. The resource demands for analysis of the data may fluctuate significantly for some of these studies (for example, seasonal or during only rush hours). Cloud computing's pay-per-use can be a preferred solution for analysing large amounts of data from these network cameras. There are few studies focusing on how to allocate cloud resources to analyse many video streams from network cameras. Existing autoscaling and load balancing methods are inapplicable due to the nature of the workloads. This paper presents a solution to manage cloud resources analysing multiple video streams. When an analysis program is launched, the resource manager adjusts the resource allocation by one of two methods: gradually scaling up the number of instances or predicting the needed number of instances. Then, the resource manager monitors the utilization and scales the resources in response to the analysis program's loads. The proposed resource manager has been implemented using Microsoft Azure and demonstrated that the system can adaptively adjust resource to cater to the workload. We explore the trade-offs inherent to both methods. Gradually scaling up can reach more efficient resource allocation, but may take multiple trials. The prediction method is faster in making allocation decisions, but may need additional adjustment. We also investigate the impact of the types of virtual machines and find that smaller sizes are usually better in terms of performance-cost ratios.

Constructive alignment has been shown to elicit higher levels of learning among students. Problem-based Learning is one of the forms of constructive alignment often used in medicine and engineering education. We have applied a PBL-based approach to a graduate (master) course in Data Intensive Systems taught simultaneously through a video link at two universities in Europe and USA. Application of standardized measuring methodology shows inconclusive impact of PBL approach on students' learning. We present survey results and analyze major factors that could have lead to inconclusive result, including: low data quality, general applicability of constructivism in computer science, and issues with hybrid classroom and alternative laboratory environments. Finally, we discuss reversed classroom method as a potential solution to the issues encountered.

This paper describes a cloud-based software infrastructure for teaching big data analytics. Using this infrastructure, students can retrieve and analyze real-time visual data retrieved from globally distributed network cameras.

Many cyber infrastructure and cloud computing systems have been developed and deployed over the past decade. Although use metrics are collected by many of these systems, there is not a clear link from these metrics to the ultimate effectiveness and impact of these systems on science communities. This paper describes a framework we developed that seeks to provide context for use and impact metrics to facilitate understanding of how these systems are used and ultimately adopted by science and engineering communities. We use this framework to present metrics of use, impact, and effectiveness collected from the NEES cyber infrastructure.

In high performance computing systems running on native hardware or cloud computing resources, parallel applications can reserve a large number of resources for long time periods. Resource failures trigger the failure of applications using this resource. Our investigation of large-scale systems in the field has revealed a difference in the operational reliability of nodes. By adding awareness of this difference to the scheduler along with the predicted reliability needs, we match reliable resources with the most demanding applications to reduce the probability of application failure. In this paper, we describe a new approach we developed to enhance reliability and reduce failure costs. Our approach partitions resources based on expected reliability and sizes each partition to bound the probability of blocking requests. Our approach can be used to size systems for peak loads with a bounded probability of blocking requests, and would be useful for operators seeking to improve the reliability and efficiency of systems.

The emergence of Big Data and Data Intensive Systems as specialized fields in computing is motivating the development of new courses and curricula. The biggest challenge is to define coherent and stable intended learning outcomes in a highly dynamic field. It is also necessary to balance between proper knowledge acquisition by student and the required time investment of teacher. We carefully looked into education theory and based our course on the problem based learning approach. We demonstrate how to define relevant intended learning outcomes, problems, teaching/learning activities and we align them with assessment tasks. Preliminary evaluation shows that intended learning outcomes are well covered by defined problems and that alignment between teaching/learning activities and assessment tasks has been reached. We hope this paper documenting our approach will be useful to other groups facing a similar task, we also think it can be helpful in establishing of Common Body of Knowledge for Data Intensive Science and related areas.

SUMMARY Network for Earthquake Engineering Simulation (NEES) is a network of 14 earthquake engineering labs distributed across the USA. As a part of the NEES effort NEESComm operates a comprehensive cyberinfrastructure that consists of the NEEShub and the NEES Project Warehouse. NEESComm provides consistent access to several high performance computing (HPC) venues. These venues include Extreme Science and Engineering Discovery Environment, the Open Science Grid, Purdue Supercomputers, and NEEShub servers. In this paper, we describe the system we developed, batchsubmit, which allows NEES researchers to make use of all these venues through the NEEShub science gateway. Copyright © 2014 John Wiley & Sons, Ltd.

The tremendous growth in data volumes has created a need for new tools and algorithms to quickly analyze large datasets. Cluster analysis techniques, such as K-means can be used for large datasets distributed across several machines. The accuracy of K-means depends on the selection of seed centroids during initialization. K-means++ improves on the K-means seeder, but suffers from problems when it is applied to large datasets: (a) the random algorithm it employs can produce inconsistent results across several analysis runs under the same initial conditions; and (b) it scales poorly for large datasets. In this paper we describe a new Competitive K-means algorithm we developed that addresses both of these problems. We describe an efficient MapReduce implementation of our new Competitive K-means algorithm that we found scales well with large datasets. We compared the performance of our new algorithm with three existing cluster analysis algorithms and found that our new algorithm improves cluster analysis accuracy and decreases variance. Our results show that our new algorithm produced a speedup of 76 ± 9 times compared with the serial K-means++ and is as fast as the Streaming K-means. Our work provides a method to select a good initial seeding in less time, facilitating accurate cluster analysis over large datasets in shorter time.

The amount of resources needed to provision Virtual Machines (VM) in a cloud computing systems to support virtual HPC clusters can be predicted from the analysis of historic use data. In previous work, Hacker et al. found that cluster analysis is a useful tool to understand the underlying spatio-temporal dependencies present in system fault and use logs. However, the cluster analysis used for reducing spatio-temporal dependences should be fast and accurate to understand the underlying stochastic properties of these systems. K-means is a fast cluster analysis method, in which accuracy depends on the use of initialization algorithms that are usually serial and slow. In this paper we present two new parallel strategies for fast seeding K-means cluster analysis. Both strategies were tested on a real problem where the aim was to reduce spatial and temporal dependencies of failures on large supercomputer systems. The performance of both strategies were compared with five existing serial implementations: K-means implementations of 1) Lloyd (L); 2) McQueen (M); and 3) Hartigan - Wong (HW), all of them using Forgy seeding; 4) K-means++; and 5) Neural Gas clustering (NG), a more recent and sophisticated method. Our results show that our new Parallel Competitive Fitness approach reduces the Within Sum of Squares (WSQQ) measure, thus increasing cluster quality of the three K-means implementations: L; M; HW, and is 200 times faster than the existing serial K-means++. The existing serial and our new Parallel K-means++ have the lowest WSQQ. Our new Parallel K-means++ is twice as fast as the existing serial K-means++ method, and is 4 times faster than the NG method. Moreover, our new methods did not generate empty clusters, while NG did. As a result of our new techniques, predicting the amount of resources needed to provision VMs processing historic system fault and use data can now be done faster and with more accuracy.

Virtualization and cloud computing technologies now make it possible to create scalable and reliable virtual high performance computing clusters. Integrating these technologies, however, is complicated by fundamental and inherent differences in the way in which these systems allocate resources to computational tasks. Cloud computing systems immediately allocate available resources or deny requests. In contrast, parallel computing systems route all requests through a queue for future resource allocation. This divergence of allocation policies hinders efforts to implement efficient, responsive, and reliable virtual clusters.

In the evolution of computing technology over the decades, file system capabilities have not grown in tandem to processing power. Today, scientific computing is highly data intensive and relies on workflows. Workflows developed are not portable among workflow management system. Also, scientific computation that rely on inherent workflows do not have a kernel support for workflows, and are executed essentially in a batch processing model. A file system that includes native kernel functionalities to support workflow execution would address the issue of parallel processing as well as portability. Such a file system would improve scientific computing performance. This paper describes an approach we developed to add workflow functionality to the Linux kernel and native file system to help simplify the use of workflow management systems for scientific computing.

Purdue University operates one of the largest cycle recovery systems in existence in academia based on the Condor workload management system. This system represents a valuable and useful cyberinfrastructure (CI) resource supporting research and education for campus and national users. During the construction and operation of this CI, we encountered many unforeseen challenges and benefits unique to an actively used infrastructure of this size. The most significant problems were integrating Condor with existing campus UPC resources, managing resource and user growth, coping with the distributed ownership of compute resources around campus, and integrating this CI with the Tera- Gridand Open Science Grid. In this paper, we describe some of our experiences and establish some best practices, which we believe will be valuable and useful to other academic institutions seeking to operate a production campus cyberinfrastrucure of a similar scale and utility.

Many high performance computing systems today exploit the availability and remarkable performance characteristics of stand alone server systems and the impressive price / performance ratio of commodity components. Small scale HPC systems, in the range from 16 to 64 processors, have enjoyed significant popularity and are an indispensable tool for the research community. Scaling up to hundreds and thousands of processors, however, has exposed operational issues, which include system availability and reliability. In this paper, we explore the impact of individual component reliability rates on the overall reliability of an HPC system. We derive a mathematical model for determining the failure rate of the system, the probability of failure of a job running on a subset of the system, and show how to design a reasonable queue structure to provide a reliable system over abroad job mix. We also explore the impact of reliability and queue structure on checkpoint intervals and recovery. Our results demonstrate that it is possible to design a reliable high performance computing system with very good operational reliability characteristics from a collection of moderately reliable components.

Applications that use parallel TCP streams to increase throughput must multiplex and demultiplex data blocks over a set of TCP streams transmitting on one or more network paths. When applications use the obvious round robin scheduling algorithm for multiplexing data blocks, differences in transmission rate between individual TCP streams can lead to significant data block reordering. This forces the demultiplexing receiver to buffer out-of-order data blocks, consuming memory and potentially causing the receiving application to stall. This paper describes a new adaptive weighted scheduling approach for multiplexing data blocks over a set of parallel TCP streams. Our new scheduling approach, compared with the scheduling approached used by GridFTP, reduces reordering of data blocks between individual TCP streams, maintains the aggregate throughput gains of parallel TCP, consumes less receiver memory for buffering out-of-order packets, and delivers smoother application goodput. We demonstrate the improved characteristics of our new scheduling approach using data transmission experiments over real and emulated wide-area networks.

Applications that require good network performance often use parallel TCP streams and TCP modifications to improve the effectiveness of TCP. If the network bottleneck is fully utilized, this approach boosts throughput by unfairly stealing bandwidth from competing TCP streams. Improving the effectiveness of TCP is easy, but improving effectiveness while maintaining fairness is difficult. In this paper, we describe an approach we implemented that uses a long virtual round trip time in combination with parallel TCP streams to improve effectiveness on underutilized networks. Our approach prioritizes fairness at the expense of effectiveness when the network is fully utilized. We compared our approach with standard parallel TCP over a wide-area network, and found that our approach preserves effectiveness and is fairer to competing traffic than standard parallel TCP.

While the advent of distributed and grid computing systems will open new opportunities for scientific exploration, the reality of such implementations could prove to be a system administrator’s nightmare. A lot of effort is being spent on identifying and resolving the obvious problems of security, scheduling, authentication and authorization. Lurking in the background, though, are the largely unaddressed issues of accountability and usage accounting:

The use of parallel TCP connections to increase throughput for bulk transfers is common practice within the high performance computing community. However, the effectiveness, fairness, and efficiency of data transfers across parallel connections is unclear. This paper considers the impact of systemic non-congestion related packet loss on the effectiveness, fairness, and efficiency of parallel TCP transmissions. The results indicate that parallel connections are effective at increasing aggregate throughput, and increase the overall efficiency of the network bottleneck. In the presence of congestion related losses, parallel flows steal bandwidth from other single stream flows. A simple modification is presented that reduces the fairness problems when congestion is present, but retains effectiveness and efficiency.