Daniela Rosu

An overview of the Rialto real-time architecture

The goal of the Rialto project at Microsoft Research is to build a system architecture supporting coexisting independent real-time (and non-real-time) programs. Unlike traditional embedded-systems real-time environments, where timing and resource analysis among competing tasks can be done off-line, it is our goal to allow multiple independently authored real-time applications with varying timing and resource requirements to dynamically coexist and cooperate to share the limited physical resources available to them, as well as also coexisting with non-real-time applications.This paper gives an overview of the Rialto real-time architecture as it is implemented today and reports on some of the early results obtained. In particular, it describes the use of time constraints, activities, CPU and other resource reservation, and the system resource planner, and how they work together to achieve our goal of providing a flexible, dynamic real-time computing environment.

FARA-a framework for adaptive resource allocation in complex real-time systems

The paper introduces FARA, a framework that provides abstractions and mechanisms for building integrated adaptation and resource allocation services in complex real time systems. FARAs design addresses: (1) the specification of the adaptation capabilities of general domain applications with multiple adaptive components; and (2) the control of the adaptation impact on the satisfiability of an applications timing constraints during its transition to targeted steady state. We propose to use of a hierarchical adaptation model, to control the adaptation enactment overhead based on the costs of executing the application specific adaptation procedures, and to reduce the decision overhead by use of effective solutions for mechanisms frequently invoked during the decision process, such as the decision context management and the allocation decision evaluation.

Multi-dimensional Knowledge Integration for Efficient Incident Management in a Services Cloud

The increasing complexity and dynamics in IT infrastructure and the emerging Cloud services present challenges to timely incident/problem diagnosis and resolution. In this paper we present a problem determination platform with multi-dimensional knowledge integration (e.g. configuration data, system vital data, log data, related tickets) and enablement for efficient incident and problem management of the enterprise. Three features of the platform are discussed: automated ticket classification, the automated association of resource with tickets based on integration with configuration database, and the collection of the system vitals relevant to the ticket through integration with monitoring systems. In response to the emerging Cloud services and their highly dynamic service operation context, we identify the need for a proactive service management approach which incorporates configurations and deployment of incident management tools, policies, and templates throughout the service life cycle in order to enable effective and efficient incident management in service operation.

An evaluation of TCP splice benefits in web proxy servers

This study is the first to evaluate the performance benefits of using the recently proposed TCP Splice kernel service in Web proxy servers. Previous studies show that splicing client and server TCP connections in the IP layer improves the throughput of proxy servers like firewalls and content routers by reducing the data transfer overheads. In a Web proxy server, data transfer overheads represent a relatively large fraction of the request processing overheads, in particular when content is not cacheable or the proxy cache is memory-based. The study is conducted with a socket-level implementation of TCP Splice. Compared to IP-level implementations, socket-level implementations make possible the splicing of connections with different TCP characteristics, and improve response times by reducing recovery delay after a packet loss. The experimental evaluation is focused on HTTP request types for which the proxy can fully exploit the TCP Splice service, which are the requests for non-cacheabl.content and SSL tunneling. The experimental testbed includes an emulated WAN environment and benchmark applications for HTTP/1.0 Web client, Web server, and Web proxy running on AIX RS/6000 machines. Our experiments demonstrate that TCP Splice enables reductions in CPU utilization of 10-43% of the CPU, depending on file sizes and request rates. Larger relative reductions are observed when tunneling SSL connections, in particular for small file transfers. Response times are also reduced by up to 1.8sec.

Improving the accessibility of aurally rendered HTML tables

Current techniques employed to aurally render HTML tables often result in output that is very difficult for sight-impaired users to understand. This paper proposes TTPML, an XML-compliant markup language, which facilitates the generation of prose descriptions of tabular information. The markup language enables content creators to specify contextual reinforcement of, and linear navigation through, tabular information. The markup language may be applied to pre-existing Web content and is reusable across multiple tables. TTPML may be interpreted by origin servers, proxy servers, or browsers. We believe that our approach benefits sight-impaired users by improving accessibility to tabular information.

Towards an optimized model of incident ticket correlation

In recent years, IT Service Management (ITSM) has become one of the most researched areas of IT. Incident and Problem Management are two of the Service Operation processes in the IT Infrastructure Library (ITIL). These two processes aim to recognize, log, isolate and correct errors which occur in the environment and disrupt the delivery of services. Incident Management and Problem Management form the basis of the tooling provided by an Incident Ticket Systems (ITS).

FARACost: an adaptation cost model aware of pending constraints

The perturbations induced by adaptation and resource allocation decisions on the adapted applications may have the undesirable side effect of causing timing constraint failures. In order to benefit from available adaptation capabilities yet avoid critical timing failures, the dynamic resource allocation mechanism should be aware of the perturbation induced by its decisions. Therefore, the impact of adaptation on short-term performance should be considered a first-class decision criterion, along with traditional criteria such as long-term performance and application criticality. Towards this end, we propose the FARACost, an adaptation cost model that captures the impact of application-specific adaptation procedures and uses this information to evaluate adaptation choices. Experimental evaluations with two applications demonstrate that the use of models like FARACost reduces or prevents pending timing constraint failures, while leading to long-term performance improvements. The experiments are conducted in a cluster environment with a fully implemented infrastructure for adaptation and resource allocation based on the FARA framework.

Optimizing Change Request Scheduling in IT Service Management

Enterprises of today face the challenge of managing large, complex IT eco-systems consisting of software applications, servers, network routers, and other type of resources. Change management, especially scheduling of changes, is known to be one of the most challenging problems in managing IT operations. In this paper, we propose an optimization model for IT change scheduling that takes into account the constraints and cost factors typically encountered in a service provider environment. In particular, we formulate the model in a way that can be solved using standard mathematical programming techniques (i.e., mixed integer programming). This not only results in strictly optimal solutions, but also provides a scalable means for scheduling a large set of change requests with complex constraints. Furthermore, having a computational efficient optimization solution facilitates the study of the scheduling sensitivity with respect to parameter inaccuracy and leads to more robust change schedules. Finally, we demonstrate the effectiveness of the proposed approach in an IT change management example which is built using insights from a large service delivery account and over two hundred thousand change instances.

Architecting Web sites for high performance

Web site applications are some of the most challenging high-performance applications currently being developed and deployed. The challenges emerge from the specific combination of high variability in workload characteristics and of high performance demands regarding the service level, scalability, availability, and costs. In recent years, a large body of research has addressed the Web site application domain, and a host of innovative software and hardware solutions have been proposed and deployed. This paper is an overview of recent solutions concerning the architectures and the software infrastructures used in building Web site applications. The presentation emphasizes three of the main functions in a complex Web site: the processing of client requests, the control of service levels, and the interaction with remote network caches.

Hint-based acceleration of Web proxy caches

Numerous studies show that proxy cache miss ratios are typically at least 40%-50%. This paper proposes and evaluates a new approach for improving the throughput of proxy caches by reducing cache miss overheads. An embedded system able to achieve significantly better communications performance than a traditional proxy filters the requests directed to a proxy cache, forwarding the hits to the proxy and processing the misses itself. This system, called a Proxy Accelerator, uses hints of the proxy cache content and may include a main memory cache for the hottest objects. Scalability with the Web proxy cluster size is achieved by using several accelerators. We use analytical models and trace-based simulations to study the benefits and the implementation tradeoffs of this new approach. A single proxy accelerator node in front of a four-node Web proxy improves the cost-performance ratio by about 40%. Implementation choices that do not affect the overall hit ratio are available.