The desire of companies to analyze web-site activity data quickly in order to show personalized content and advertisements to users has led to renewed interest in continuous query processing. One important query class here is windowed aggregation which does time-based windowing followed by grouping and aggregation over a data stream. An example query may aggregate each user's activity over a recent one hour window, and update the result every five minutes. In this paper, we characterize the rich execution plan space for windowed aggregation queries. No such attempt has been made previously to the best of our knowledge. Our second contribution is in developing a cost-based optimizer to pick a good plan from this space for a given query. Finally, we show the effectiveness of the cost-based optimizer.

As the data collected by enterprises grows in scale, there is a growing trend of performing data analytics on large datasets. Batch processing systems that can handle petabyte scale of data, such as Hadoop, have flourished and gained traction in the industry. As the results of batch analytics have been used to continuously improve front-facing user experience, there is a growing interest in pushing the processing latency down. This trend has fueled a resurgence in the development and usage of execution engines that can process continuous queries. An important class of continuous queries is windowed aggregation queries. Such queries arise in a wide range of applications such as generating personalized content and results. Today, considerable manual effort goes into finding the most suitable execution engine for these queries and on tuning query performance on these engines. An ecosystem composed of multiple execution engines may be needed in order to run the overall query workload efficiently given the diverse set of requirements that arise in practice. Cyclops is a continuous query processing platform that manages and orchestrates windowed aggregation queries in an ecosystem composed of multiple continuous query execution engines. Cyclops employs a cost-based approach for picking the most suitable engine and plan for executing a given query. This demonstration first presents an interactive visualization of the rich execution plan space of windowed aggregation queries, which allows users to analyze and understand the differences among plans. The next part of the demonstration will drill down into the design of Cyclops. For a given query, we show the cost spectrum of query execution plans across three different execution engines---Esper, Storm, and Hadoop---as estimated by Cyclops.

Background: The Agency for Integrated Care (AIC) conducted a survey to determine the specific areas in which the agency could work with General Practitioners (GPs) to enhance mental illness management in Singapore. This survey was conducted as part of AIC’s engagement strategy and because the Primary Care Surveys 2010 and 2005 noted that a sizeable number of mental health cases are seen by GPs. Methodology: An interviewer-administered questionnaire was conducted among a random sample of 849 group and non-group GPs practicing in four geographical districts of Singapore. The response rate was 64%. Findings: About 70% of the respondent GPs indicated that they were seeing patients with mental illness. Anxiety and depression were the most commonly seen mental conditions. The lack of available drugs and facilities in the clinic, and unfamiliarity with mental health patient management were the main reasons that GPs do not see patients with mental illness. About 39% of the GPs referred patients to a psychologist. About 36% of the GPs referred patients to a counsellor. Availability of additional allied health services in the neighbourhood was crucial in encouraging GPs to see more patients with mental illness. General geriatrics, dementia, general mental health and major depression were the top four areas of interest for further training. Conclusions: The role of GPs in managing mental illnesses remains crucial. There is more pressing need to address their perceived difficulties in managing such patients. The results of the GP Landscape Survey provided evidential bases to enhance the National Mental Health Blueprint in three areas: (i) extension of GP consultation subsidies to middle-income patients, (ii) expansion of allied mental health services to support GPs, and (iii) development of training programmes for GPs in mental illnesses that are commonly encountered by and of interest to GPs.

no fits all is a sound philosophy for system de- signers to follow, it poses multiple challenges for application de- velopers and system administrators. It can be hard for an applica- tion developer to pick system when the needs of her applica- tion match the features of multiple one size systems. The choice becomes considerably harder when different components of an ap- plication fit the features of different one size systems. Consider- able manual effort goes into creating and tuning such multi-system applications. An application's data and workload properties may change over time, often in unpredictable and bursty ways. Conse- quently, the one size system that is best for an application can change over time. Adapting to change can be hard when applica- tion development is coupled tightly with any individual one size system. In this paper, we make the case for developing a new breed of Database Management Systems that we term DBMS + . A DBMS + contains multiple one size systems internally. An application specifies its execution requirements on aspects like performance,

Globally, healthcare systems have been largely centered on acute care characterized by episodic, highlevel (often expensive) care with sophisticated medical sub-specialization. There is now consensus globally that there should be a fundamental refocus by healthcare systems towards more care outside of the acute setting so as to deal with the challenges of an aging population and increased prevalence of chronic diseases. Singapore has a resident population of some 5.5 million people and is also one of the fastest aging populations in the world today. This presents tremendous challenges and opportunities for designing and implementing different care delivery models that meet the needs of an aging population.

There is a growing trend of performing analysis on large datasets using workflows composed of MapReduce jobs connected through producer-consumer relationships based on data. This trend has spurred the development of a number of interfaces--ranging from program-based to query-based interfaces--for generating MapReduce workflows. Studies have shown that the gap in performance can be quite large between optimized and unoptimized workflows. However, automatic cost-based optimization of MapReduce workflows remains a challenge due to the multitude of interfaces, large size of the execution plan space, and the frequent unavailability of all types of information needed for optimization. We introduce a comprehensive plan space for MapReduce workflows generated by popular workflow generators. We then propose Stubby, a cost-based optimizer that searches selectively through the subspace of the full plan space that can be enumerated correctly and costed based on the information available in any given setting. Stubby enumerates the plan space based on plan-to-plan transformations and an efficient search algorithm. Stubby is designed to be extensible to new interfaces and new types of optimizations, which is a desirable feature given how rapidly MapReduce systems are evolving. Stubby's efficiency and effectiveness have been evaluated using representative workflows from many domains.

Timely and cost-effective analytics over “Big Data” is now a key ingredient for success in many businesses, scientific and engineering disciplines, and government endeavors. The Hadoop software stack—which consists of an extensible MapReduce execution engine, pluggable distributed storage engines, and a range of procedural to declarative interfaces—is a popular choice for big data analytics. Most practitioners of big data analytics—like computational scientists, systems researchers, and business analysts—lack the expertise to tune the system to get good performance. Unfortunately, Hadoop’s performance out of the box leaves much to be desired, leading to suboptimal use of resources, time, and money (in payas-you-go clouds). We introduce Starfish, a self-tuning system for big data analytics. Starfish builds on Hadoop while adapting to user needs and system workloads to provide good performance automatically, without any need for users to understand and manipulate the many tuning knobs in Hadoop. While Starfish’s system architecture is guided by work on self-tuning database systems, we discuss how new analysis practices over big data pose new challenges; leading us to different design choices in Starfish.

Online social networks (OSNs) are immensely popular, but their centralized control of user data raises important privacy concerns. This paper presents Vis-à-Vis, a decentralized framework for OSNs based on the privacy-preserving notion of a Virtual Individual Server (VIS). A VIS is a personal virtual machine running in a paid compute utility. In Vis-à-Vis, a person stores her data on her own VIS, which arbitrates access to that data by others. VISs self-organize into overlay networks corresponding to social groups. This paper focuses on preserving the privacy of location information. Vis-à-Vis uses distributed location trees to provide efficient and scalable operations for sharing location information within social groups. We have evaluated our Vis-à-Vis prototype using hundreds of virtual machines running in the Amazon EC2 compute utility. Our results demonstrate that Vis-à-Vis represents an attractive complement to today's centralized OSNs.

Power costs are very significant for data centers. To maximally utilize the provisioned power capacity, data centers often employ over-subscription, that is, the sum of peak consumptions of individual servers may be greater than the provisioned capacity. Power budgeting methods are employed to ensure that actual consumption never exceeds capacity. However, current power budgeting methods enforce capacity limits in hardware and are not well suited for virtualized servers because the hardware is shared among multiple applications. We present a power budgeting system for virtualized infrastructures that enforces power limits on individual distributed applications. Our system enables multiple applications to share the same servers but operate with their individual quality of service guarantees. It responds to workload and power availability changes, by dynamically allocating appropriate amount of power to different applications and tiers within applications. The design is mindful of practical constraints such the data center's limited visibility into hosted application performance. We evaluate the system using workloads derived from real world data center traces.

Elasticity - where systems acquire and release resources in response to dynamic workloads, while paying only for what they need - is a driving property of cloud computing. At the core of any elastic system is an automated controller. This paper addresses elastic control for multi-tier application services that allocate and release resources in discrete units, such as virtual server instances of predetermined sizes. It focuses on elastic control of the storage tier, in which adding or removing a storage node or "brick" requires rebalancing stored data across the nodes. The storage tier presents new challenges for elastic control: actuator delays (lag) due to rebalancing, interference with applications and sensor measurements, and the need to synchronize the multiple control elements, including rebalancing. We have designed and implemented a new controller for elastic storage systems to address these challenges. Using a popular distributed storage system - the Hadoop Distributed File System (HDFS) - under dynamic Web 2.0 workloads, we show how the controller adapts to workload changes to maintain performance objectives efficiently in a pay-as-you-go cloud computing environment.

Online social networks (OSNs) are immensely popular, but their centralized control of user data raises important privacy concerns. This paper presents Vis-„a-Vis, a decentralized framework for OSNs based on the privacy-preserving notion of a Virtual Individual Server (VIS). A VIS is a personal virtual machine running within a paid compute utility. In Vis-„a-Vis, a person stores her location data on her own VIS, which arbitrates access to that data by others. VISs self-organize into overlay networks corresponding to social groups with whom their owners wish to share location information. Vis-„a-Vis uses distributed location trees to provide efficient and scalable operations for creating, joining, leaving, searching, and publishing data to these groups. We have evaluated our Vis-„a-Vis prototype in the Amazon EC2 compute utility. Our results demonstrate that Vis-„a-Vis represents an attractive complement to today’s centralized OSNs.

People increasingly generate content on their mobile devices and upload it to third-party services such as Facebook and Google Latitude for sharing and backup purposes. Although these services are convenient and useful, their use has important privacy implications due to their centralized nature and their acquisitions of rights to user-contributed content. This paper argues that people's interests would be be better served by uploading their data to a machine that they themselves own and control. We term these machines Virtual Individual Servers (VISs) because our preferred instantiation is a virtual machine running in a highly-available utility computing infrastructure. By using VISs, people can better protect their privacy because they retain ownership of their data and remain in control over the software and policies that determine what data is shared with whom. This paper also describes a range of applications of VIS proxies. It then presents our initial implementation and evaluation of one of these applications, a decentralized framework for mobile social services based on VISs. Our experience so far suggests that building such applications on top of the VIS concept is feasible and desirable.

With advances in virtualization technology, virtual machine services offered by cloud utility providers are becoming increasingly powerful, anchoring the ecosystem of cloud services. Virtual computing services are attractive in part because they enable customers to acquire and release computing resources for guest applications adaptively in response to load surges and other dynamic behaviors. ``Elastic'' cloud computing APIs present a natural opportunity for feedback controllers to automate this adaptive resource provisioning, and many recent works have explored feedback control policies for a variety of network services under various assumptions. This paper addresses the challenge of building an effective controller as a customer add-on outside of the cloud utility service itself. Such external controllers must function within the constraints of the utility service APIs. It is important to consider techniques for effective feedback control using cloud APIs, as well as how to design those APIs to enable more effective control. As one example, we explore proportional thresholding, a policy enhancement for feedback controllers that enables stable control across a wide range of guest cluster sizes using the coarse-grained control offered by popular virtual compute cloud services.

Online social networks (OSN) are hugely popular, but their centralized control of sensitive user data raises important privacy concerns. This paper presents Vis-à-Vis, a decentralized framework for online social networking based on the privacy-preserving notion of a Virtual Individual Server (VIS), a personal virtual machine running within a cloud-computing utility. In Vis-à-Vis, each person manages personal data such as friend lists, photographs, and messages through his own VIS. In addition, VISs collectively self-organize into larger groups corresponding to shared user attributes such as interests and educational background. Vis-à-Vis uses distributed hash tables to provide a decentralized, efficient, and scalable set of operations for joining, leaving, and searching a wide range of OSN groups. We have evaluated our Vis-à-Vis prototype implementation in several virtualcomputing environments using experimental parameters observed in the Facebook OSN. Our results show that our prototype’s performance is reasonable and demonstrate that Vis-à-Vis represents an attractive and feasib le alternative to today’s centralized OSNs.

Mobile social services enable increasingly popular forms o f ocial interaction among mobile-device users, such as finding out w hen people of interest are nearby. However, today’s services ra ise important privacy concerns because they concentrate locatio n inf rmation for many users under a single administrative domain. We present a privacy-preserving framework for these services in which each person maintains her own location history in her own Vir tual Individual Server (VIS), a personal virtual machine ru nning in a utility computing infrastructure. We enable location s haring among VISs through self-organizing overlay networks, one p er social group with which VIS owners wish to share information. W e describe an implementation of this framework that exploits skip graphs and Z-order space-filling curves to provide efficient a d scalable operations on distributed location data. Finally , we demonstrate the feasibility of our approach by evaluating our imp lementation using real-world location traces gathered for this pur pose. Our decentralized approach makes large-scale privacy breache s much less likely than in centralized architectures, and gives pe ople fine control over what location information they share with whom .