Electrical computers and digital processing systems: multicomput – Computer network managing – Computer network monitoring
Reexamination Certificate
1999-12-28
2003-01-07
Luu, Le Hien (Department: 2752)
Electrical computers and digital processing systems: multicomput
Computer network managing
Computer network monitoring
C702S182000, C702S186000
Reexamination Certificate
active
06505249
ABSTRACT:
BACKGROUND
1. Field of the Invention
The present invention relates to client-server based computer systems, and more particularly, to a method for benchmarking and optimizing the end-to-end processing performance of a client-server based computer system to determine the optimal values of the system variables.
2. Background of the Invention
Virtually every large business today relies on a computer network system to manage information. Typically, these systems operate multiple software applications to handle business functions such as accounting, purchasing, inventory, work scheduling, and customer service. The businesses invest huge sums of money to develop custom networks that efficiently meet these specific functions. In return for such large investments, the businesses demand the fastest and most reliable systems possible.
Conscious of these demands, system designers attempt to predict system demands, such as the number and frequency of transactions or the number of concurrent users. Based on these predictions, the designers assemble hardware components capable of meeting the anticipated demands. As apparent to one skilled in the art, the process is often little more than an initial educated guess followed by trial and error. Considering the large number of applications a system is typically asked to support, this unfocused approach is seemingly unavoidable because of the sheer number of hardware and software variables involved in assembling a system.
Faced with such a large number of variables, system designers must first focus on making the system work. Unfortunately, multiple components manufactured by different companies and the applications layered on top of them introduce a limitless number of possible programming problems. By trial and error, system designers adjust relationships between hardware and software until a workable model is found. Often, ensuring proper communication between components and accomplishing overall system operation are the designers' only goals. Thus, designers are usually satisfied just to resolve programming problems and achieve system functionality, without even considering system efficiency.
If the system designers do set aside time and effort for system optimization, they typically focus on improving the performance of the individual system components. On the component level, designers can more easily understand and evaluate the variables affecting performance. In addition, the system designers can draw from the experiences and recommendations of individual component manufacturers. For instance, a hardware manufacturer, such as Hewlett Packard™, knows in detail the factors that improve or degrade hardware performance. Database engineers understand completely the optimal conditions for efficient performance of a database, such as Oracle™. Thus, typically designers optimize the performance of each of the components, incorporate the optimized components into the system, and expect that the overall system is operating at close to its most efficient capability. However, because these components depend heavily on each other and on the specific application of the system, this assumption is often incorrect. Plainly stated, optimized components do not always add up to an optimized system.
The strategy of component optimization is a relic of earlier monolithic processing environments. Monolithic refers to a network system designed, manufactured, and supported by a single manufacturer. In the early days of network computing, a single manufacturer supplied the network, was responsible for overall end-to-end processing performance and, as the designer of each component, had the knowledge of the individual components and the overall system to meet optimization needs. However, as the computer industry advanced and increased in complexity, manufacturers began to specialize and produce only a single component of a network. These manufacturers still furnished design and optimization support for customers, but only for their respective components. Thus, today, the customer, who is typically the resident network engineer of a business, is burdened with the responsibility of assembling optimized components that do not necessarily amount to an optimized network system.
SUMMARY OF THE INVENTION
The present invention is a method for evaluating and optimizing the performance of a client-server based computer network system. Specifically, the present invention provides a method for benchmarking the end-to-end processing performance of a computer network system comprised of individually manufactured components. The benchmarking method identifies performance-affecting system variables, measures their relative impacts on performance, and arrives at optimal settings for overall system optimization.
As most broadly defined, benchmarking is a way to measure the efficiency and effectiveness of computer system performance. Typically, benchmarking is used as part of wide scale information technology metrics, which is defined as the science of measuring the efficiency and effectiveness of computer hardware, software, and systems. Once a system is operational, an initial benchmark test is run in which the system performs the function for which it was designed. This benchmark test evaluates the performance of the system by some measurable unit, such as time or frequency of performance cycles. In addition, the benchmark test notes the respective values of the system variables.
As described above, these benchmarking tests are typically applied only to components of a computer network. As opposed to this component level strategy, the present invention assesses performance on a system-wide level, taking into account system component interrelationships. The present invention is primarily intended for actual computer network systems that are in use, and not necessarily to model systems. The present invention analyzes computer network systems devoted to particular applications, and identifies different system destabilizers for each system and each application. The present invention provides the following method of system performance benchmarking, which can be performed either manually, by system designers, or preferably automatically, by software executing the steps described herein.
First, a set of performance variables that impact system performance is identified. These performance variables, also referred to as system destabilizers, influence the performance of the software, hardware, and network processes. Examples of these variables include the number of central processing units (CPUs), the number of concurrent users, and the network protocols. Once the destabilizers are identified, a software matrix is constructed to represent and track the performance variables. Next, the functional (but not optimized) system is performance-tested. This initial performance test establishes the baseline for the system, which is recorded in the software matrix.
With a baseline established, the performance variables are pinned to constant values except for one variable, which is allowed to float through different values. While this one variable is floating, system performance is recorded in the software matrix, including metrics pertinent to the particular software application, such as processing speed, system access and congestion, and system response time. Readings for each value of the floating variable are recorded as the values change through successive thresholds.
Once a statistically complete sample has been taken, the floating variable is returned to its baseline value and the method is repeated with a new performance variable. A test sample is statistically complete when the test results demonstrate that a relationship between the floating variable and system performance does or does not exist, and demonstrates such a relationship to the degree of accuracy required by the specific application of the computer network system. After tracking the computer network's response to changes in value of the different system destabilizers, the present invention presents a clear picture of
BellSouth Intellectual Property Corporation
Luu Le Hien
Shaw Pittman LLP
LandOfFree
Method for optimizing end-to-end processing performance by... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method for optimizing end-to-end processing performance by..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for optimizing end-to-end processing performance by... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3012338