Search

We study the business-driven prioritization of workflows in enterprise transaction processing systems. Today, IT services are increasingly shared amongst workflows, and static prioritization is applied during phases of high system workload to mitigate the business impact of service queuing and resulting workflow response-time increase. Static prioritization rules can only be optimal for one particular workflow definition and demand mix scenario, while in practice workflow definitions are continuously re-engineered and demand varies over time. We propose a technique to automatically adapt workflow priorities aimed at minimizing overload-related costs resulting from workflow delays. We continuously align maximum service buffer lengths and bid-prices reflecting current marginal opportunity costs of invoking a shared service. New prioritization is derived by comparing expected costs when queuing a workflow with its current opportunity costs, computed as the sum of bid-prices of invoked services. Simulation outcomes based on industry data illustrate that the approach dominates standard queuing disciplines as well as static prioritization.

We consider the assignment of enterprise applications in virtual machines to physical servers, also known as server consolidation problem. Data center operators try to minimize the number of servers, but at the same time provide sufficient computing resources at each point in time. While historical workload data would allow for accurate workload forecasting and optimal allocation of enterprise applications to servers, the volume of data and the large number of resulting capacity constraints in a mathematical problem formulation renders this task impossible for any but small instances. We use singular value decomposition (SVD) to extract significant features from a large constraint matrix and provide a new geometric interpretation of these features, which allows for allocating large sets of applications efficiently to physical servers with this new formulation. While SVD is typically applied for purposes such as time series decomposition, noise filtering, or clustering, in this paper features are used to transform the original allocation problem into a low-dimensional integer program with only the extracted features in a much smaller constraint matrix. We evaluate the approach using workload data from a large data center and show that it leads to high solution quality, but at the same time allows for solving considerably larger problem instances than what would be possible without data reduction and model transform. The overall approach could also be applied to similar packing problems in service operations management.

IT service providers are increasingly hosting different services of different customers on a shared IT infrastructure. While this fosters utilization of hardware infrastructure, system malfunctions, unexpected service behaviour or peak demands for one or more services may exploit resource pools (CPU, I/O, main memory, bandwidth etc.), entailing rejection of service requests. In this paper we describe models for dynamic admission control on shared infrastructures. The admission control model decides whether to accept, buffer or reject a service request based on the revenue, Service Level Agreements (SLAs) and its resource demand in comparison to the actual workload to maximize overall revenue. Simulations of a media streaming infrastructure have been used for evaluation and comparison with traditional admission control policies.

Prioritization of business processes in distributed transaction processing systems is applied to circumvent drawbacks resulting from queueing effects. These effects arise because of increasing workload of business processes using shared services. To predict the performance of such systems, typically queueing and simulation models are used and prioritization schemes are created for specific business process definitions and demand mixes. Unfortunately, the derived solution may get obsolete after every demand-change and process re-engineering cycle. We propose an approach to adapt the prioritization by continuously aligning threshold-prices for the service usage reflecting the opportunity costs of using a service. The prioritization of a process is then reduced by the sum of current threshold prices of services invoked by that process. First we show that this approach is asymptotically optimal with an increasing number of business processes and workload requests. Second, we conduct simulations to show that the dynamic prioritization approach dominates common static prioritization approaches even in scenarios with moderate numbers of processes and process requests.

Virtualized data centers are hosting virtual machines (VMs) running enterprise applications with time varying resource demand behavior (workload) jointly on the same physical servers in order to increase server utilization. To avoid server overload situations, a data center operator needs to decide which VMs should be assigned to a physical server for a given period of time. As assignment decisions are based on historical workload traces, deviations from forecasted demands potentially require the adaptation of VM assignments. However, VM live migrations are inherently overhead afflicted control operations. While it is mandatory to anticipate overload situations in order to trigger VM migrations in a proactive manner, unnecessary VM migrations may have negative impact on the underlying computing infrastructure. Hence, an autonomic controller should accurately predict situations where the aggregated workload of a set of collocated VMs will hit the capacity limit of a server without requiring manual adjustments of control model parameters. In this paper we propose an automated, non-parametric approach for proactive filtering of multivariate workload behavior. We learn an orthonormal projection from historical workload traces and extract a set of key metrics that concisely describe relevant developments in the joint workload behavior of physical servers. A geometric interpretation, in combination with simple short term forecasting techniques allows for reliable decision making. Based on a set of real world workload traces we conduct numerical experiments that validates its superiority and predictive capabilities over simple threshold based approaches.

We introduce a novel method to discover beneficial time frames for adapting virtual machine (VM) assignments in consolidated enterprise data centers. Our key insight lies in learning an optimal orthonormal transform from the workload data of a set of enterprise applications hosted in VMs. The transform allows us to extract only a few indicators from long, time-varying and complex workload time series. The indicators represent the initially high-dimensional data set in a reduced form which allows for a concise identification of periods of relatively stable resource demands and turning points in the behavior of a set of VM workloads that require VM reassignments. In this work, we address one of the most pressing problems for data center operators, namely the reduction of managerial complexity of resource and workload management in data centers hosting thousands of applications with complex and varying workload behaviors. We demonstrate the decision support model using workload traces from a professional data center.1

Server virtualization enables dynamic workload management for data centers. However, especially live migrations of virtual machines (VM) induce significant overheads on physical hosts and the shared network infrastructure possibly leading to host overloads and SLA violations of co-hosted applications. While some recent work addresses the impact of live migrations on CPUs of physical hosts, little attention has been given to the control and optimization of migration algorithms and migration-related network bandwidth consumption. In this paper we introduce network topology aware scheduling models for VM live migrations. We propose a scheme for classifying VMs based on their workload characteristics and propose adequate resource and migration scheduling models for each class, taking network bandwidth requirements of migrations and network topologies into account. We also underline the necessity for additional migration control parameters for efficient migration scheduling.