A dynamic server migration and consolidation algorithm is introduced. The algorithm is shown to provide substantial improvement over static server consolidation in reducing the amount of required capacity and the rate of service level agreement violations. Benefits accrue for workloads that are variable and can be forecast over intervals shorter than the time scale of demand variability. The management algorithm reduces the amount of physical capacity required to support a specified rate of SLA violations for a given workload by as much as 50% as compared to static consolidation approach. Another result is that the rate of SLA violations at fixed capacity may be reduced by up to 20%. The results are based on hundreds of production workload traces across a variety of operating systems, applications, and industries.

http://ants.iis.sinica.edu.tw/3BkMJ9lTeWXTSrrvNoKNFDxRm3zFwRR/91/Dynamic%20Placement%20of%20Virtual%20Machines.pdf

Dynamic Placement of Virtual Machines for Managing SLA Violations

Journal of the ACM

The underutilization of disk parallelism and file cache buffers by traditional file systems induces I/O stall time that degrades the performance of modern microprocessor-based systems. In this paper, we present aggressive mechanisms that tailor file system resource management to the needs of I/O-intensive applications. In particular, we show how to use application-disclosed access patterns (hints) to expose and exploit I/O parallelism and to allocate dynamically file buffers among three competing demands: prefetching hinted blocks, caching hinted blocks for reuse, and caching recently used data for unhinted accesses. Our approach estimates the impact of alternative buffer allocations on application execution time and applies a cost-benefit analysis to allocate buffers where they will have the greatest impact. We implemented informed prefetching and caching in DEC''s OSF/1 operating system and measured its performance on a 150 MHz Alpha equipped with 15 disks running a range of applications including text search, 3D scientific visualization, relational database queries, speech recognition, and computational chemistry. Informed prefetching reduces the execution time of the first four of these applications by 20% to 87%. Informed caching reduces the execution time of the fifth application by up to 30%.

/pdf/informed-prefetching-and-caching-4nm5o41nvm.pdf

Informed prefetching and caching

How does a search engine company decide what ads to display with each query so as to maximize its revenueq This turns out to be a generalization of the online bipartite matching problem. We introduce the notion of a trade-off revealing LP and use it to derive an optimal algorithm achieving a competitive ratio of 1−1/e for this problem.

/pdf/adwords-and-generalized-online-matching-13dglrv5l6.pdf

AdWords and generalized online matching

The delivery of video content is expected to gain huge momentum, fueled by the popularity of user-generated clips, growth of VoD libraries, and wide-spread deployment of IPTV services with features such as CatchUp/PauseLive TV and NPVR capabilities. The `time-shifted' nature of these personalized applications defies the broadcast paradigm underlying conventional TV networks, and increases the overall bandwidth demands by orders of magnitude. Caching strategies provide an effective mechanism for mitigating these massive bandwidth requirements by replicating the most popular content closer to the network edge, rather than storing it in a central site. The reduction in the traffic load lessens the required transport capacity and capital expense, and alleviates performance bottlenecks. In the present paper, we develop light-weight cooperative cache management algorithms aimed at maximizing the traffic volume served from cache and minimizing the bandwidth cost. As a canonical scenario, we focus on a cluster of distributed caches, either connected directly or via a parent node, and formulate the content placement problem as a linear program in order to benchmark the globally optimal performance. Under certain symmetry assumptions, the optimal solution of the linear program is shown to have a rather simple structure. Besides interesting in its own right, the optimal structure offers valuable guidance for the design of low-complexity cache management and replacement algorithms. We establish that the performance of the proposed algorithms is guaranteed to be within a constant factor from the globally optimal performance, with far more benign worst-case ratios than in prior work, even in asymmetric scenarios. Numerical experiments for typical popularity distributions reveal that the actual performance is far better than the worst-case conditions indicate.

/pdf/distributed-caching-algorithms-for-content-distribution-5car9o0t6h.pdf

Distributed Caching Algorithms for Content Distribution Networks

Speed scaling is a power management technique that involves dynamically changing the speed of a processor. We study policies for setting the speed of the processor for both of the goals of minimizing the energy used and the maximum temperature attained. The theoretical study of speed scaling policies to manage energy was initiated in a seminal paper by Yao et al. [1995], and we adopt their setting. We assume that the power required to run at speed s is P(s) = sα for some constant α > 1. We assume a collection of tasks, each with a release time, a deadline, and an arbitrary amount of work that must be done between the release time and the deadline. Yao et al. [1995] gave an offline greedy algorithm YDS to compute the minimum energy schedule. They further proposed two online algorithms Average Rate (AVR) and Optimal Available (OA), and showed that AVR is 2α − 1 αα-competitive with respect to energy. We provide a tight αα bound on the competitive ratio of OA with respect to energy.We initiate the study of speed scaling to manage temperature. We assume that the environment has a fixed ambient temperature and that the device cools according to Newton's law of cooling. We observe that the maximum temperature can be approximated within a factor of two by the maximum energy used over any interval of length 1/b, where b is the cooling parameter of the device. We define a speed scaling policy to be cooling-oblivious if it is simultaneously constant-competitive with respect to temperature for all cooling parameters. We then observe that cooling-oblivious algorithms are also constant-competitive with respect to energy, maximum speed and maximum power. We show that YDS is a cooling-oblivious algorithm. In contrast, we show that the online algorithms OA and AVR are not cooling-oblivious. We then propose a new online algorithm that we call BKP. We show that BKP is cooling-oblivious. We further show that BKP is e-competitive with respect to the maximum speed, and that no deterministic online algorithm can have a better competitive ratio. BKP also has a lower competitive ratio for energy than OA for α ≥5.Finally, we show that the optimal temperature schedule can be computed offline in polynomial-time using the Ellipsoid algorithm.

Speed scaling to manage energy and temperature

We introduce and evaluate a middleware clustering technology capable of allocating resources to web applications through dynamic application instance placement. We define application instance placement as the problem of placing application instances on a given set of server machines to adjust the amount of resources available to applications in response to varying resource demands of application clusters. The objective is to maximize the amount of demand that may be satisfied using a configured placement. To limit the disturbance to the system caused by starting and stopping application instances, the placement algorithm attempts to minimize the number of placement changes. It also strives to keep resource utilization balanced across all server machines. Two types of resources are managed, one load-dependent and one load-independent. When putting the chosen placement in effect our controller schedules placement changes in a manner that limits the disruption to the system.

/pdf/dynamic-placement-for-clustered-web-applications-132drokyc2.pdf

Dynamic placement for clustered web applications

We first consider online speed scaling algorithms to minimize the energy used subject to the constraint that every job finishes by its deadline. We assume that the power required to run at speed s is P(s) = s/sup /spl alpha//. We provide a tight /spl alpha//sup /spl alpha// bound on the competitive ratio of the previously proposed optimal available algorithm. This improves the best known competitive ratio by a factor of 2/sup /spl alpha//. We then introduce an online algorithm, and show that this algorithm's competitive ratio is at most 2(/spl alpha//(/spl alpha/ - 1))/sup /spl alpha//e/sup /spl alpha//. This competitive ratio is significantly better and is approximately 2e/sup /spl alpha/+1/ for large /spl alpha/. Our result is essentially tight for large /spl alpha/. In particular, as /spl alpha/ approaches infinity, we show that any algorithm must have competitive ratio e/sup /spl alpha// (up to lower order terms). We then turn to the problem of dynamic speed scaling to minimize the maximum temperature that the device ever reaches, again subject to the constraint that all jobs finish by their deadlines. We assume that the device cools according to Fourier's law. We show how to solve this problem in polynomial time, within any error bound, using the ellipsoid algorithm.

Dynamic speed scaling to manage energy and temperature

In this paper we consider an optimization problem which models the dynamic placement of applications on servers under two simultaneous resource requirements: one that is dependent on the loads placed on the applications and one that is independent. The demand (load) for applications changes over time and the goal is to satisfy all the demand while changing the solution (assignment of applications to servers) as little as possible. We describe the system environment where this problem arises, present a heuristic algorithm to solve it, and provide an experimental analysis comparing the algorithm to previously known algorithms. The experiments indicate that the new algorithm performs much better. Our algorithm is currently deployed in the IBM flagship product Websphere.

Dynamic application placement under service and memory constraints

High-performance I/O systems depend on prefetching and caching in order to deliver good performance to applications. These two techniques have generally been considered in isolation, even though there are signi cant interactions between them; a block prefetched too early reduces the e ectiveness of the cache, while a block cached too long reduces the effectiveness of prefetching. In this paper we study the effects of several combined prefetching and caching strategies for systems with multiple disks. Using disk-accurate tracedriven simulation, we explore the performance characteristics of each of the algorithms in cases in which applications provide full advance knowledge of accesses using hints. Some of the strategies have been published with theoretical performance bounds, and some are components of systems that have been built. One is a new algorithm that combines the desirable characteristics of the others. We nd that when performance is limited by I/O stalls, aggressive prefetching helps to alleviate the problem; that more conservative prefetching is appropriate when signi cant I/O stalls are not present; and that a single, simple strategy is capable of doing both.

Tracy Kimbrel

Papers

Speed scaling to manage energy and temperature

Dynamic placement for clustered web applications

Dynamic speed scaling to manage energy and temperature

Dynamic application placement under service and memory constraints

A Trace-Driven Comparison of Algorithms for Parallel Prefetching and Caching (CMU-CS-96-174)