Akhil Sahai

Correct synchronization among activities is critical in a business process. Current process languages such as BPEL specify the control flow of processes

The characterization of distributed n-tier application performance is an important and challenging problem due to their complex structure and the significant variations in their workload. Theoretical models have difficulties with such wide range of environmental and workload settings. Experimental approaches using manual scripts are error-prone, time consuming, and expensive. We use code generation techniques and tools to create and run the scripts for large-scale experimental observation of n-tier benchmarking application performance measurements over a wide range of parameter settings and software/hardware combinations. Our experiments show the feasibility of experimental observations as a sound basis for performance characterization, by studying in detail the performance achieved by (up to 3) database servers and (up to 12) application servers in the RUBiS benchmark with a workload of up to 2700 concurrent users.

Utility services providers offer on-demand services for multiple users and applications from a shared resource pool. The utility computing environment provider manages resources that have multiple dimensions, namely CPU, Memory and Disk. It is important that a mechanism of quantification he utilized for determining pricing. Also in these environments, we believe that price needs to be adapted to reflect and influence dynamic change in demand on resources, and can be used, as an economic mechanism, to efficiently manage a utility computing environment. To achieve this, we propose in this paper an adaptive pricing mechanism. The objective of this scheme is twofold: it, for business objectives, adapts price based upon the supply and demand curves over time intervals; on the other hand, price invisibly manages the distribution of resource demand across time intervals, therefore makes resource utilization more efficient and balanced over time. In this approach, the resource unit to be priced is called capacity unit, and applications are billed based upon the capacity units that have been reserved or used.

Shared services, hosted in utility computing environments are becoming quite common in enterprises. These services could be clusters of application servers hosting various applications or even clusters of databases catering to multiple organizations within an enterprise. Automating the provisioning of these shared services is a difficult problem. We propose a model-based automation mechanism to provision such shared services. We demonstrate our model-based provisioning system in the case of a shared desktop service. The shared desktop service termed the "virtual desktop system" (VDS) hosts desktops on top of virtual machines. These desktop virtual machines are provisioned on consolidated data centers while users use thin clients to connect to their "virtual" desktops. We demonstrate how we use our model-based automation technique to simplify the provisioning of such desktops.

Utility computing environments provide on-demand IT services to customers. Such environments are dynamic in nature and continuously adapt to changes in requirements and system state. Errors are an important category of environment state changes as such environments consist of a large number of components, and hence, are subject to errors. In this paper, we design and implement a model-based simulation framework that leverages information about existing service components and their interactions, and provides concrete service behavior in presence of a variety of errors. To evaluate the framework, experiments are conducted on a virtualized blade-server based environment. Results show that the framework is effective and practical in analyzing error impacts on IT services.

Systems are typically designed based on certain high level goals, such as performance and availability. On the other hand, during operation, usually only low level metrics (e.g., CPU utilization) are measured. The system administrators use their expertise to implicitly map bounds on these metrics such that the high level goals are met. The objective of this research is to create an automated and domain independent approach to derive policy bounds on the low level metrics such that the high level goals are met. These policies may be also be used for monitoring the system for goal assessment purposes. The refinement is carried out using a combination of data classification and test-and-development approaches. An ad hoc system is deployed and a dataset containing values of selected metrics is collected by placing appropriate workloads on the system. The policy bounds are derived by applying classification techniques on this dataset. The classification rules are further refined using statistical distributions to arrive at certain low level rules that are useful for system monitoring and to check the system health when it is deployed and running. We show the validity of our approach for an e-commerce auctioning system (RubiS).

The current approach for modeling synchronization in business processes relies on sequencing constructs, such as sequence, parallel etc. However, sequencing constructs obfuscate the true source of dependencies in a business process. Moreover, because of the nested structure and scattered code that results from using sequencing constructs, it is hard to add or delete additional constraints without over-specifying necessary constraints or invalidating existing ones. We propose a dataflow programming approach in which dependencies are explicitly modeled to guide activity scheduling. We first give a systematic categorization of dependencies: data, control, service and cooperation. Each dimension models dependency from its own point of view. Then we show that dependencies of various kinds can be first merged and then optimized to generate a minimal dependency set, which guarantees high concurrency and minimal maintenance cost for process execution.

In this work, the performance of various machine learning classifiers with regard to bottleneck detection in enterprise, multi-tier applications governed by service level objectives is described. Specifically, in this paper, it demonstrates the effectiveness of three classifiers, a tree-augmented Naive Bayesian network, a J48 decision tree, and LogitBoost, using our bottleneck detection process, which delves into a new area of performance analysis based on the trends of metrics (first order derivative) rather than the metric value itself. Furthermore, the efficiency of each classifier by measuring the convergence speed, or the number of staging trials required in order to provide positive results is illustrated. Finally, the effectiveness of the classifiers used in the bottleneck detection process as each classifier strongly identifies the enterprise system bottleneck