5,086 research outputs found
Unifying Access and Resource Usage Control over Standard Client-Server Interactions
No full textWe propose a novel framework for integrated access and resource usage control over standard clientserver interactions. Historically, access control has been developed without considering resource usage. Resource control has thus developed as an ad hoc server-centric set of mechanisms (e.g., file system quota, network bandwidth quote, etc.). We believe that resource usage control is strongly related to access control and so should be implemented using a unified, global enforcement framework. We introduce such a framework, where services have resource usage constraints and principals have resource usage histories. To access and use a service, a principal must have the appropriate access and sufficient resource usage rights when considering its usage history. Our framework is able to enforce global stateful policies, yet do not require changes to existing message-passing applications. We have built a prototype and used it to specify and enforce an example policy that includes role-based control and delegation. We applied our system to control access and resource usage for three different services, network, DNS, and SMB file systems, to demonstrate its effectiveness and wide applicability.Technical report DCS-TR-67
Examining Race/Ethnicity Diversity in the Enrollment of a 4-year CS University Program
No full textComputing departments in the United States are not producing the number of graduates that the workforce needs. At the same time, they lack gender and racial diversity. Therefore, it is important to attract students from minority groups that comprise a significant part of the US population and can bring an enriching, diverse perspective to the development of new technology. In this paper, we study patterns of enrollment and retention among minority students in a 4-year CS university program in order to better understand the challenges for increasing racial and ethnic diversity. We use student data from three core CS classes with a special focus on the introduction to programming (CS1) class. We compare the ethnic makeup of our CS enrollment with that of the student population at our university, that of the state and with nationwide numbers, and examine how the ethnic gap changes from an introductory programming class to an upper level class. We also analyze how different factors such as intent to major, prior experience in computing, and CS1 grades correlate with student retention
Unified Structure and Content Search for Personal Information Management Systems
No full textThe amount of data that users are storing and accessing in personal information systems is growing massively. At the same time, the organization of this data is becoming more heterogeneous, with data spread across different organizational domains such as emails, music databases, and photo albums, some of which are structured by applications rather than users. Powerful search tools are needed to help users locate data in these rapidly expanding yet fragmented data sets. In this paper, we present a novel fuzzy search approach that considers approximate matches to structure and content query conditions. Our approach includes a scoring framework for computing unified relevance scores for potential answers. Critically, our framework uses unified data and query processing models so that structure conditions can be approximately matched by content inside files and vice versa. Our model also unifies external structure (directories) with internal structure (e.g., XML structure), allowing users to specify integrated queries that are matched to a single unified data domain. We propose indexes and algorithms for efficient query processing. Finally, we empirically evaluate our approach using a real data set. We show that our unified fuzzy search approach can leverage structure information to significantly improve search accuracy, yet is robust to mistakes in query conditions.Technical report DCS-TR-66
MassConf: Automatic Configuration Tuning By Leveraging User Community Information
No full textConfiguring modern enterprise software can be extremely difficult, because its behavior often depends on large numbers of configuration parameters. We argue that vendors can simplify the configuration process for new users of their software by collecting and using configuration information from the existing user community. Our proposed approach is based on the observations that (1) a “good” configuration may work well for many different users, and (2) multiple good configurations may work well for each user. We demonstrate our idea by designing MassConf, a system that collects and uses existing configurations to automatically configure new software installations. To evaluate MassConf, we use it to configure the Apache Web server to achieve a response-time target. Our results confirm our observations and show that MassConf successfully reaches the targets of many more new installations than an existing efficient optimization algorithm. Even when we consider only the installations that can be configured with this efficient algorithm, our results show that MassConf reaches the desired targets running many fewer experiments on average.Technical report DCS-TR-66
Quantifying and Improving I/O Predictability in Virtualized Systems
No full textVirtualization enables the consolidation of virtual machines (VMs) to increase the utilization of physical servers in Infrastructure-as-a-Service (IaaS) cloud providers. Unfortunately, our quantification of storage I/O performance across a range of workloads, virtual machine monitor (VMM) architectures, approaches to storage virtualization, and storage devices shows widespread performance unpredictability in the face of consolidation. Surprisingly, the use of solid-state drives (SSDs) can exacerbate the problem. Since many users may desire consistent performance, we argue that IaaS cloud providers should provide a class of predictable-performance service in addition to their existing (predictability-oblivious) services. Thus, we propose and evaluate VirtualFence, a storage system that provides predictable performance for this new class of service. VirtualFence uses three main techniques: (1) non-work-conserving time-division I/O scheduling, (2) a small SSD cache in front of a much larger hard disk drive (HDD), and (3) space-partitioning of both the SSD cache and the HDD. Our evaluation of a prototype VirtualFence implemented in the Xen VMM shows that VirtualFence improves predictability significantly. More fundamentally, our evaluation illustrates the tradeoff between predictability and performance. We conclude that current VMMs are far from providing predictability. Systems like VirtualFence can remedy this problem, while allowing the cloud provider to select an appropriate compromise between performance and predictability.Technical report DCS-TR-69
Efficient Multi-Dimensional Query Processing in Personal Information Management Systems
No full textThe relentless growth in capacity and dropping price of storage are driving an explosion in the amount of information users are collecting and storing in personal information management systems. This explosion of information has led to a critical need for complex search tools to access often very heterogeneous data in asimple and efficient manner. Such tools should provide both high-quality flexible scoring mechanisms and efficient query processing capabilities. In this paper, we focus on indexes and algorithms to efficiently identify the most relevant files that match multi-dimensional queries comprised of relaxed content, metadata, and structure conditions. We also adapted existing top-k query strategies to our specific scenario. Our work is integrated in Wayfinder, an existing fully functioning file system. We perform a thorough experimental evaluation of our file search techniques and show that our query processing strategies exhibit good behavior across all dimensions, resulting in good overall query performance and good scalability.Technical report DCS-TR-62
GreenCassandra: Using Renewable Energy in Distributed Structured Storage Systems
No full textOn-site generation of renewable (“green”) energy can help to significantly reduce grid (“brown”) energy consumption, and correspondingly the carbon footprint, of datacenters. However, it is challenging to use green energy generated from sources such as solar and wind because energy production is variable. In this paper, we investigate how to manage an interactive service, where response time is a critical performance metric, to maximize the benefits of green energy produced from these sources. Specifically, we design, prototype, and evaluate a distributed structured storage system, GreenCassandra, which is representative of a class of important subsystems underlying many interactive cloud services. Our proposed approach predicts the production of solar energy, and then controls the number of active nodes to manage energy consumption while respecting a response time SLA. When green energy is available, GreenCassandra may activate extra servers to build up slack with respect to the SLA. When using brown energy, GreenCassandra deactivates servers to reduce energy consumption, leveraging any built-up performance slack, while observing constraints imposed by the SLA. Evaluations show that GreenCassandra can use a heuristic green-energy-aware policy to decrease brown energy consumption and cost by up to 28% and 29%, respectively. Further, these savings are very close to those achievable by an optimizationbased policy that has perfect knowledge of future workload and green energy production.Technical report DCS-TR-71
ApproxHadoop: Bringing Approximations to MapReduce Frameworks
No full textResearch has shown that approximate computing is effective at reducing the resource requirements, computation time, and/or energy consumption of large-scale computing. In this paper, we We propose and evaluate a framework for creating and running approximation-enabled MapReduce programs. Specifically, we propose approximation mechanisms that fit naturally into the MapReduce paradigm, including input data sampling, task dropping, and accepting and running a precise and a user-defined approximate version of the MapReduce code. We then show how to leverage statistical theories to compute error bounds for popular classes of MapReduce programs when approximating with input data sampling and/or task dropping. We implement the proposed mechanisms and error bound estimations in a prototype system called ApproxHadoop. Our evaluation uses MapReduce applications from different domains, including data analytics, scientific computing, video encoding, and machine learning. Our results show that ApproxHadoop can significantly reduce application execution time and/or energy consumption when the user is willing to tolerate small errors. For example, ApproxHadoop can reduce runtimes by up to 32× when the user can tolerate an error of 1% with 95% confidence. We conclude that our framework and system can make approximation easily accessible to many application domains using the MapReduce model.Technical report DCS-TR-70
Cost-Aware Building of Cloud Services Powered by Renewable Energy
No full textInterest in powering datacenters at least partially using on-site renewable sources, e.g. solar or wind, has been growing. In fact, researchers have studied distributed services comprising networks of such “green” datacenters, and load distribution approaches that “follow the renewables” to maximize their use. Unfortunately, the prior works have not considered where to site such a network for efficient production of renewable energy, while minimizing both datacenter and renewable plant building costs. Moreover, researchers have not built real load management systems for follow-the-renewables services. Thus, in this paper, we propose a framework, optimization problem, and solution approach for siting and provisioning green datacenters for a follow-the-renewables HPC cloud service. We illustrate the location selection tradeoffs by quantifying the minimum cost of achieving different amounts of renewable energy. Finally, we design and implement a system capable of migrating virtual machines across the green datacenters to follow the renewables. Among other interesting results, we demonstrate that one can build green HPC cloud services at a relatively low additional cost compared to existing servicesTechnical report DCS-TR-70
GreenSlot: Scheduling Energy Consumption in Green Datacenters
No full textIn this paper, we propose GreenSlot, a parallel batch job scheduler for a datacenter powered by a photovoltaic solar array and the electrical grid (as a backup). GreenSlot predicts the amount of solar energy that will be available in the near future, and schedules the workload to maximize the green energy consumption while meeting the jobs’ deadlines. If brown energy must be used to avoid deadline violations, the scheduler selects times when brown electricity is cheap. Our results for production scientific workloads demonstrate that GreenSlot can increase green energy consumption by up to 117% and decrease energy cost by up to 39%, compared to EASY backfilling. Based on these positive results, we conclude that green datacenters and green-energy-aware scheduling can have a significant role in building a more sustainable IT ecosystem.Technical report DCS-TR-68
- …
