Research Interests:
Research Interests:
www.elsevier.com/locate/peva Performance impacts of autocorrelated flows in multi-tiered systems
Research Interests:
Storage systems are often deployed in a tiered form to enable high performance and availability. These tiers utilize all possible volatile and non-volatile storage technologies, including DRAM, SSD, and HDD. The tradeoffs among their... more
Storage systems are often deployed in a tiered form to enable high performance and availability. These tiers utilize all possible volatile and non-volatile storage technologies, including DRAM, SSD, and HDD. The tradeoffs among their cost, features, and capabilities can make their effective integration into a single storage entity complex. Here, we propose an autonomic technique that learns user traffic patterns in a storage system over long time-scales to optimize user performance but also volume of completed system work. Our purpose is to multiplex as best as possible user workload with storage system features (e.g., voluminous internal system work) such that the latter is not starved but rather completed with minimal impact on user performance. Key to achieving the above is to use an autonomic learning engine to predict when the user workload intensity increases/decreases and then proactively stop/start bulky internal system work. Being proactive allows the system to effectively ...
Research Interests:
Research Interests:
Research Interests:
Low disk drive utilization suggests that placing the drive into a power saving mode during idle times may decrease power consumption. We present PREFiguRE, a robust framework that aims at harvesting future idle intervals for power savings... more
Low disk drive utilization suggests that placing the drive into a power saving mode during idle times may decrease power consumption. We present PREFiguRE, a robust framework that aims at harvesting future idle intervals for power savings while meeting strict quality constraints: first, it contains potential delays in serving IO requests that occur during power savings since the time to bring up the disk is not negligible, and second, it ensures that the power saving mechanism is triggered a few times only, such that the disk wear-out due to powering up and down does not compromise the disk’s lifetime. PREFiguRE is based on an analytic methodology that uses the histogram of idle times to determine schedules for power saving modes as a function of the preceding constraints. PREFiguRE facilitates analysis for the evaluation of the trade-offs between power savings and quality targets for the current workload. Extensive experimentation on a set of enterprise storage traces illustrates P...
Research Interests:
The biggest power consumer in data centers is the storage system. Coupled with the fact that disk drives are lowly utilized, disks offer great opportunities for power savings, but any power saving action should be transparent to user... more
The biggest power consumer in data centers is the storage system. Coupled with the fact that disk drives are lowly utilized, disks offer great opportunities for power savings, but any power saving action should be transparent to user traffic. Estimating correctly the performance impact of power saving becomes crucial for the effectiveness of power saving. Here, we develop a methodology that quantitatively estimates the performance impact due to power savings. By taking into consideration the propagation delay effects. Experiments driven by production server traces verify the correctness and efficiency of the proposed analytical methodology.
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
As storage in data centers is increasing rapidly, it has become critical to find ways to operate efficiently this important component of a data center. Often, it has been proposed to consolidate the storage workload into a subset of... more
As storage in data centers is increasing rapidly, it has become critical to find ways to operate efficiently this important component of a data center. Often, it has been proposed to consolidate the storage workload into a subset of storage devices and shutdown the unused ones with the purpose of preserving power. In many cases storage workload consolidation requires some amount of data to be copied from one device or set to the next. While storage workload consolidation techniques focus on extending power savings with minimal penalty in the performance of a data center, less attention is paid to the process of seamlessly integrating the data copy phase into the overall storage workload consolidation technique. Specifically, in this paper, we propose an analytic framework that synchronizes the pace of copying data between two storage devices (or nodes) such that performance is maintained within predefined targets. As such, we avoid either undesired performance degradation caused by ...
Research Interests:
Research Interests:
Research Interests:
Although recent advances in theory indicate that burstiness in the service time process can be handled effectively by queueing models (e.g.,MAP queueing networks [2]), there is a lack of understanding and of practical results on how to... more
Although recent advances in theory indicate that burstiness in the service time process can be handled effectively by queueing models (e.g.,MAP queueing networks [2]), there is a lack of understanding and of practical results on how to perform model parameterization, especially when this parameterization must be derived from limited coarse measurements. We propose a new parameterization methodology based on the index of dispersion of the service process at a server, which is inferred by observing the number of completions within the concatenated busy periods of that server. The index of dispersion together with other measurements that reflect the "estimated" mean and the 95th percentile of service times are used to derive a MAP process that captures well burstiness of the true service process. Detailed experimentation on a TPC-W testbed where all measurements are obtained via a commercially available tool, the HP (Mercury) Diagnostics, shows that the proposed technique off...
Research Interests:
Research Interests:
Size-based policies have been known to successfully bal- ance load and improve performance in homogeneous clus- ter environments where a dispatcher assigns a job to a server strictly based on the job size. We first examine how size-based... more
Size-based policies have been known to successfully bal- ance load and improve performance in homogeneous clus- ter environments where a dispatcher assigns a job to a server strictly based on the job size. We first examine how size-based policies can provide service differentiation and complement admission control and/or priority scheduling policies. We find that under autocorrelated arrivals the ef- fectiveness
Research Interests:
Research Interests: Mathematics, Applied Mathematics, Computer Science, Technology, Performance Evaluation, and 10 moreComputation, Mathematical Sciences, State Space, Computational Efficiency, Exact Computation, Markov chain, Numerical Stability, Linear Equations, System of Equation Linear Equations, and Direct Method
Temporal dependence, as a synonym for burstiness, is often found in workloads (i.e., arrival flows and/or service times) in enterprise systems that use the multi-tier paradigm. Despite the fact that burstiness has deleterious effects on... more
Temporal dependence, as a synonym for burstiness, is often found in workloads (i.e., arrival flows and/or service times) in enterprise systems that use the multi-tier paradigm. Despite the fact that burstiness has deleterious effects on performance, existing modeling and benchmarking techniques do not provide an effective capacity planning for multi-tier systems with temporal dependence. In this paper, we first present strong evidence that existing models cannot capture bursty conditions and accurately predict performance. Therefore, we propose a simple and effective sizing methodology to integrate workload burstiness into models and benchmarking tools used in system sizing. This modeling methodology is based on the index of dispersion which jointly captures variability and burstiness of the service process in a single number. We report experimentation on a real testbed that validates the accuracy of our modeling technique by showing that experimental and model prediction results ar...
Research Interests:
Research Interests:
Research Interests:
Research Interests: Computer Science, Distributed Computing, Software Engineering, Queueing theory, Computer Hardware, and 11 moreOPERATING SYSTEM, Performance Analysis, Complex System, Computer Software, Performance Prediction, State Space, Capacity Planning, Model Generation, Markov chain, System performance, and Performance Index
Research Interests:
Burstiness and temporal dependence in service processes are often found in multi-tier architectures and storage devices and must be captured accurately in capacity planning models as these features are responsible of significant... more
Burstiness and temporal dependence in service processes are often found in multi-tier architectures and storage devices and must be captured accurately in capacity planning models as these features are responsible of significant performance degradations. However, existing models and approximations for networks of first-come first-served (FCFS) queues with general independent (GI) service are unable to predict performance of systems with temporal dependence in workloads. To overcome this difficulty, we define and study a class of closed queueing networks where service times are represented by Markovian Arrival Processes (MAPs), a class of point processes that can model general distributions, but also temporal dependent features such as burstiness in service times. We call these models MAP queueing networks. We introduce provable upper and lower bounds for arbitrary performance indexes (e.g., throughput, response time, utilization) that we call Linear Reduction (LR) bounds. Numerical ...
Research Interests:
Various processor allocation policies for multiprogrammed multiprocessor systems have beenproposed in the literature. The focus of this paper is on the class of non-preemptive policies,since they represent a viable choice for... more
Various processor allocation policies for multiprogrammed multiprocessor systems have beenproposed in the literature. The focus of this paper is on the class of non-preemptive policies,since they represent a viable choice for implementation in actual systems. Several allocationpolicies are evaluated from the perspectives of both performance and applicability.A uniform comparison of policies is given based on the parameters required for implementation.A
The broadening disparity between the performance of I/O devices and the performanceof processors and communication links on parallel platforms is a major obstacleto achieving high performance in many parallel application domains. We... more
The broadening disparity between the performance of I/O devices and the performanceof processors and communication links on parallel platforms is a major obstacleto achieving high performance in many parallel application domains. We believe thatunderstanding the interactions among application I/O access patterns, parallel file systems,and I/O hardware configurations is a prerequisite to identifying levels of I/Oparallelism (i.e., the number of disks
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
In multiprocessor systems, a reasonable goal of the scheduler is to keep all processors as busyas possible. One technique for doing this is to allocate all available processors to the jobs waitingfor service. Techniques which allocate all... more
In multiprocessor systems, a reasonable goal of the scheduler is to keep all processors as busyas possible. One technique for doing this is to allocate all available processors to the jobs waitingfor service. Techniques which allocate all available processors are known as work-conservingpolicies. In this paper, non-work-conserving policies are examined. These policies keep somenumber of processors idle (i.e., unallocated) even
Research Interests:
Page 1. KPC-Toolbox: Best Recipes Toward Automatization of Workload Fitting ∗ Eddy Zheng Zhang College of William & Mary Computer Science Dept. Williamsburg, 23187, VA, US eddy@cs.wm.edu Giuliano Casale College of William & Mary... more
Page 1. KPC-Toolbox: Best Recipes Toward Automatization of Workload Fitting ∗ Eddy Zheng Zhang College of William & Mary Computer Science Dept. Williamsburg, 23187, VA, US eddy@cs.wm.edu Giuliano Casale College of William & Mary Computer Science Dept. ...
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Abstract. The broadening disparity in the performance of input/output (I/O) devices and the performance of processors and communication links on parallel systems is a major obstacle to achieving high perfor-mance for a wide range of... more
Abstract. The broadening disparity in the performance of input/output (I/O) devices and the performance of processors and communication links on parallel systems is a major obstacle to achieving high perfor-mance for a wide range of parallel applications. I/O hardware and file system ...