International Business Machines Corporation

Robert Krauthgamer

Jayram S Thathachar

Maxim I Sviridenko

Tracy Jay Kimbrel

Baruch M Schieber

Maria Minkoff

Nimrod Megiddo

Vijayshankar Raman

Tapas Kanungo

Ziv Bar-Yossef

Atri Rudra

Parikshit Gopalan

Aranyak Mehta

Methods are provided for maximizing the throughput of a computer system in the presence of one or more power constraints. Throughput is maximized by repeatedly or continuously or periodically optimizing task scheduling and assignment for each of a plurality of components of a computer system. The components include a plurality of central processing units (CPUs) each operating at a corresponding operating frequency. The components also include a plurality of disk drives. The corresponding operating frequencies of one or more CPUs of the plurality of CPUs are adjusted to maximize computer system throughput under one or more power constraints. Optimizing task scheduling and assignment, as well as adjusting the corresponding operating frequencies of one or more CPUs, are performed by solving a mathematical optimization problem using a first methodology over a first time interval and a second methodology over a second time interval longer than the first time interval. The first methodology comprises a short term heuristic solver for adapting to computer system changes that occur on a short time scale, and the second methodology comprises a long term solver having greater accuracy and greater computational complexity than the first methodology.

Computer system management and throughput maximization in the presence of power constraints

The embodiments of the invention provide a method of ordering an intersecting of a group of lists into a left-deep AND-tree. The method begins by performing a first selecting process including selecting a top list, corresponding to a top leaf of the left-deep AND-tree, from the group of lists to leave remaining lists of the group of lists. The top list can be the smallest list of the group of lists. The method can also select a pair of lists from the group of lists, such that the pair of lists has the smallest intersection size relative to other pairs of lists of the group of lists. Next, the method estimates intersections of the remaining lists with the top list by estimating an amount of intersection between the remaining lists and the top list. This involves sampling a portion of the remaining lists. The method also includes identifying larger list pairs having smaller intersections sizes when compared to smaller list pairs having larger intersections sizes.

Adaptive greedy method for ordering intersecting of a group of lists into a left-deep AND-tree

Disclosed is a method of scanning a data stream in a single pass to obtain uniform data samples from selected intervals. The method comprises randomly selecting elements from the stream for storage in one or more data buckets and, then, randomly selecting multiple samples from the bucket(s). Each sample is associated with a specified interval immediately prior to a selected point in time. There is a balance of probabilities between the selection of elements stored in the bucket and the selection of elements included in the samples so that elements scanned during the specified interval are included in the sample with equal probability. Samples can then be used to estimate the degree of sortedness of the stream, based on counting how many elements in the sequence are the rightmost point of an interval such that majority of the interval's elements are inverted with respect to the interval's rightmost element.

Method of obtaining data samples from a data stream and of estimating the sortedness of the data stream based on the samples

A benefit task system implements a policy for allocating resources to yield some benefit. The method implemented may be applied to a variety of problems, and the benefit may be either tangible (e.g., profit) or intangible (e.g., customer satisfaction). In one example, the method is applied to server allocation in a Web site server 'farm' given full information regarding future loads to maximize profits for the Web hosting service provider. In another example, the method is applied to the allocation of telephone help in a way to improve customer satisfaction. In yet another example, the method is applied to distributed computing problem where the resources to be allocated are general purpose computers connected in a network and used to solve computationally intensive problems. Solution of the Web server 'farm' problem is based on information regarding future loads to achieve close to the greatest possible revenue based on the assumption that revenue is proportional to the utilization of servers and differentiated by customer class. The method of server allocation uses an approach which reduces the Web server farm problem to a minimum-cost network flow problem, which can be solved in polynomial time. Similar solutions are applicable to other resource allocation problems.

Dynamic resource allocation using known future benefits

A focused random walk system produces samples of on-topic pages from a collection of hyper-linked pages such as Web pages. The focused random walk system utilizes a focused random walk to produce a focused sample, which is a random sample of Web pages focused on a topic. The focused random walk system uniformly samples pages iteratively, where each iteration follows a random link from a union of the in-links and out-links of a page. The system then classifies this randomly selected link to determine whether the page is on-topic. The random walk sampling process could comprise a hard-focus method that selects only on-topic pages at each step of the focused random walk, or a soft-focus method that allows limited divergence to off-topic pages.

System, method, and service for using a focused random walk to produce samples on a topic from a collection of hyper-linked pages

Methods are provided for maximizing the throughput of a computer system in the presence of one or more power constraints. Throughput is maximized by repeatedly or continuously optimizing task scheduling and assignment for each of a plurality of components of a computer system. The components include a plurality of central processing units (CPUs) each operating at a corresponding operating frequency. The components also include a plurality of disk drives. The corresponding operating frequencies of one or more CPUs of the plurality of CPUs are adjusted to maximize computer system throughput under one or more power constraints. Optimizing task scheduling and assignment, as well as adjusting the corresponding operating frequencies of one or more CPUs, are performed by solving a mathematical optimization problem using a first methodology over a first time interval and a second methodology over a second time interval longer than the first time interval. The first methodology comprises a short term heuristic solver for adapting to computer system changes that occur on a short time scale, and the second methodology comprises a long term solver having greater accuracy and greater computational complexity than the first methodology.

A method for server allocation in a Web server 'farm' is based on limited information regarding future loads to achieve close to the greatest possible revenue based on the assumption that revenue is proportional to the utilization of servers and differentiated by customer class. The method of server allocation uses an approach of 'discounting the future'. Specifically, when the policy faces the choice between a guaranteed benefit immediately and a potential benefit in the future, the decision is made by comparing the guaranteed benefit value with a discounted value of the potential future benefit. This discount factor is exponential in the number of time units that it would take a potential benefit to be materialized. The future benefits are discounted because by the time a benefit will be materialized, things might change and the algorithm might decide to make another choice for a potential (even greater) benefit.

Growing community of inventors

Albany, CA, United States of America

Robert Krauthgamer