D. Ferrari, G. Serazzi, and A. Zeigner. Measurement and Tuning of Computer Systems.Prentice-Hall, 1983.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....While in the early days of computing (70s) the typical systems were used in batchorinteractive mode, static workload models [9] could adequately represent the user behaviour. In the 80s, dynamic workload models were introduced, whichwere able to representvariabilities in user behavior ([4, 11, 14]) Within the last years, generativeworkload models [8, 24] have been proposed as a suitable method for bridging the gap between the user s application oriented view of the load and the actual load (physical, resource oriented requests) submitted to the system. Ihavecontributed to this field ....

D. Ferrari, G. Serazzi, and A. Zeigner. Measurement and Tuning of Computer Systems. Prentice-Hall, 1983.

....a small number of groups of similar requests. This technique is useful in driving ca 114 pacity planning studies where system saturation points are predicted. The term workload designates all the processing requests submitted to a system by the user community during any given period of time [19]. Workload characterization is the process of partitioning the global workload of a computer system into smaller sets of workload components which are in turn composed of transactions or jobs that have similar characteristics and assigning numbers that represent their typical resource demand ....

D. Ferrari, G. Serazzi, A. Zeigner, "Measurement and Tuning of Computer Systems," Prentice Hall, 1983.

....for efficient monitoring in all kinds of systems. The PSpec notion of a log is general enough to fit many existing monitoring systems with little work. For examples of recent work on monitoring, see [3, 12, 20, 24, 28, 32, 33, 38, 43] General discussions of monitor design may be found in [11] and [18] and in other textbooks on performance measurement. Performance characterization includes work on defining performance metrics, on characterizing workloads, and on techniques for determining values In a system with hard or soft real time constraints a performance assertion failure ....

....analysis has formed the basis for the design of the solver and can help specification writers to understand how to set up experiments to obtain useful results from the solver. Many textbooks on performance include discussions of performance characterization or linear regression, for example, [10, 11, 18, 23]. 1.3.3 Complementary Performance Work PSpec is an approach to performance testing and debugging based on automatically checking expectations captured in performance assertions. Other work in performance testing and debugging has taken different approaches. Profiling tools, such as gprof [14] ....

[Article contains additional citation context not shown here]

Domenico Ferrari, Giuseppe Serazzi, and Alessandro Zeigner. Measurement and Tuning of Computer Systems. Prentice-Hall, Inc., Englewood Cliffs, New Jersey 07632, 1983.

....can be performed, e.g. the instruction stream cannot be analysed without having a compiler available which in turn needs access to source code. Substantial analysis has been carded out for traditional workload characterisation which are used in capacity planning and performance tuning studies [14, 21, 8]. Workload characterisation can only be performed when all components of the system are in place, and thus all measurements can be obtained. However, in the case of the characterisation of parallel systems most performance studies are required before all the system components are available, or ....

D. Ferrari, G. Serazzi, and A. Zeigner, Measurement and Tuning of Computer Systems (PrenticeHall, New Jersey, 1983).

....workload. The accuracy of the workload model can be determined by analyzing the difference between the performance values achieved using the model and the performance values of the real workload. In a more general way, the accuracy of the workload characterization process can be stated as follows [Ferrari]: A workload model W is a perfect representation of the real workload R if the performance measures P obtained when running W and R in the same system are identical. Figure 4 Representativeness of a workload model [Menasc] 19 3.8 Compactness In general, the real workload of a computer system is ....

Ferrari, D., Serazzi, G. and Zeigner, A. (1983). Measurement and Tuning of Computer Systems, Prentice Hall, Englewood Cliffs, N.J.

....the use of the three feathers. 2.3 Performance Evaluation Techniques There are many techniques that can be used in evaluating performance. Usually these techniques are classified into three categories: measurement, simulation, and analytical modeling [Brewer91; Jain91] Some authors (e.g. Ferrari83] classifies simulation and analytical modeling into one, called modeling technique. This is because whether we use simulation or analytical modeling, we must first construct a model for the system under study. simulation is considered to be one powerful method to solve complex models ....

....parameters (or workload features) mean the measured quantities, service requests, or resource demands, which are used to model or characterize the workload. Examples are transaction types, packet sizes, page reference pattern, and so on [Jain91] 4.2. 2 Steps of Workload Characterization Ferrari [Ferrari83] groups the main operations to be performed in implementing workload models into three phases: formulation phase, construction phase, and validation phase, each of which involves several operations. They are summarized as follows. In the formulation phase, decisions should be made on what ....

[Article contains additional citation context not shown here]

Ferrari, D., Serazzi, G. and Zeigner, A., Measurement and Tuning of Computer Systems, Prentice-Hall, Englewood Cliffs, NJ, 1983.

....Simulation, as we shall see in chapter 5, shares many of the phases in the design process. It can also be used to validate, explain, and help correct errors of design in operational systems. Simulation can be used for tuning and upgrade exercise which are a necessity in the products life time [Ferrari,83] To complement the modelling advantages of a discrete event simulation we also provide the means of performing analytical analysis. The analytical models module consists of a set of queueing theory models for local area networks and distributed systems [Reiser,82] This module would execute ....

Ferrari, D., Serazzi, G. and Zeigner, A., Measurement and Tuning of Computer Systems, Prentice-Hall, Englewood Cliffs, NJ, (1983).

....as all the estimated or observed performance results will depend on the load, that the system under test has to process. While workload modeling for uniprocessor systems is a well established discipline, only a few techniques applicable in the analysis of parallel systems have been proposed (e.g. [8], 2] or see also a brief survey in [4] In this paper a systematic approach for workload modeling is proposed based on workload components, characteristics, and parameters. This approach is a framework for the construction of any type of test workload (executable and non executable) which can ....

D. Ferrari, G. Serazzi, and A. Zeigner. Measurement and Tuning of Computer Systems. Prentice-Hall, 1983.

.... of the performance evaluation study and the evaluation technique highly determines the right level of abstraction and the type of workload model (i.e. executable models in performance measurements and nonexecutable models in modeling) This observation has already been made for sequential systems [2] and is even more relevant in the analysis of PPS. An evaluation based on a high level abstraction might not be able to capture all performance relevant parameters while a representation at the lowest level might result in untractable large models. We will discuss briefly several objectives for ....

D. Ferrari, G. Serazzi, and A. Zeigner. Measurement and Tuning of Computer Systems. Prentice-Hall, Englewood Cliffs, New Jersey, 1983.

....Cycle The decision upon which type of test workload or which level of characterization to use can only be made when considering the objectives and restrictions of the whole performance evaluation study. By recalling the basic steps in a performance evaluation cycle, as described for example in [7], the influence of the performance evaluation framework on the definition and construction of the test workload will be discussed. How are the workload parameters obtained and quantified Is the quality of the model satisfying What is the objective of the analysis What are the relevant ....

D. Ferrari, G. Serazzi, and A. Zeigner. Measurement and Tuning of Computer Systems. Prentice-Hall, 1983.

....of our estimates and attempting to improve upon them. THE HIERARCHICAL DECOMPOSITION METHOD Hierarchical decomposition reduces the complexity of the model itself by identifying parts of the system that can be studied in isolation and replaced by simpler equivalent components (Courtois 1977, Ferrari et al. 1983). This technique has been widely used in computer systems performance evaluation to solve queuing models too complex to be directly tractable. It can also be applied to the evaluation of the availability of replicated data objects (Pris 1991) Second Ethernet A First Ethernet C B G Figure ....

Ferrari, D., G. Serazzi and A. Zeigner, Measurement and Tuning of Computer Systems, Prentice-Hall, Englewood Cliffs, NJ (1983).

....are no historical data to which reference can be made, it is necessary to build a probabilistic model of the new application using a graph representation of routine, module and subsystem programs which can be reduced by suitable reduction techniques. See Beizer (1978) for the principles used and Ferrari et al. (1983) for an application of the model. The process begins with expressing a program as a flow chart and a graph with nodes and links. Node elimination follows and then the graph reduction procedure is implemented. The estimated workload for the new computerised hospital information system can be ....

Ferrari, D., Serazzi, G. and Zeigner, A., (1983), Measurement and Tuning of Computer Systems, Prentice-Hall.

....Figure 3: UCATS Configuration Detail and Environment 5. Simulation Project ADSM has a variety of parameters that can be tuned to improve performance. In addition, basic throughput analysis can be used to help size the disk caches and buffers in the system. We are developing an HSM simulator [13,14,15,16] that will enable us to explore the impact of cache and buffer sizes, number of servers and placement of servers on a variety of job mixes. Working with the simulator is helping us to understand key issues of system behavior, such as workload. Our initial simulations are small; the strategy is to ....

Ferrari, D., Serazzi, G., and Zeigner, A., Measurement and Tuning of Computer Systems, PrenticeHall, Englewood Cliffs, New Jersey, 1983.

....skills and subjective judgments. However, as commented in [24] automatic clustering techniques should be combined with expert knowledge about the case being studied, in order for the clustering results to be valid. The major problems of the general approach of workload clustering are discussed in [11]. These are: the selection of the clustering algorithm and of the clustering distance metric, the appropriate scaling and normalization of the input data, and the efficient handling of large sized inputs. The behavior of real workloads is very complex and difficult to reproduce. A model for such ....

D. Ferrari, G. Serazzi, and A. Zeigner. "Measurement and Tuning of Computer Systems". Prentice-Hall, New Jersey, 1983.

....of any computer system, including software tools, are frequently done with the help of carefully constructed test workloads. These test workloads are used to measure performance criteria and to compare different systems and are referred to as benchmarks. As described by Jain [49] and Ferrari [29] the test workloads can be of two types: real workloads and synthetic workloads. Real workloads are basically the actual programs that will be used on the system under test. This generally considered to be the ideal approach, but it is sometimes unrealistic. The real application may not be ....

....characterize the actual workload, ffl build a synthetic workload model based on the parameters, and ffl develop a synthetic workload generator based on the parameters and the model. Identifying the parameters that characterize the workload is called workload characterization. Ferrari and Serrazi [28,29] and Calzarossa [19] have identified the process and techniques of workload characterization. The process involves an analysis of the workload and modeling the static and dynamic behavior of the real load. Calzarossa [19] also identifies the basic parameters for a parallel system that are similar ....

Ferrari, D., Serazzi, G., and Zeigner, A. Measurement and Tuning of Computer Systems. Prentice-Hall, 1983.

.... wide variety of users (students, faculty, research staff, clerical staff, and adminstrators) and goals (term papers, programming assignments, research work, personnel records, grades, and financial management) there is a strong need for detailed user profiles and application type characteristics [FSZ83]. Thus an extensive focus study was conducted, in 198687 [Macr87] on computer usage by students, faculty, and administrative staff to determine how their computer time was broken down by application type, and what use of campus networks they were making or planning to make. Subsequently, network ....

D.Ferrari, G.Serazzi, A.Zeigner, Measurement and Tuning of Computer Systems, Prentice Hall, Inc., Englewood Cliffs, New Jersey, 1983.

....the prevalent load patterns. This section reviews current approaches to load index design, discusses their deficiencies, and presents our approach based on automated learning. 2.1. Existing Methods Many existing methods for computing load indices use simplified queuing models of computer systems [3, 4, 12]. Almost all implemented systems use a function known as UNIX style load average (hereafter, load average) which is an exponentially smoothed average of the total number of processes competing for CPU. Load average meaningfully compares loading situations across configurationally identical ....

....function but rather in other parameters of the load balancing policy [8] Let us examine whether the right hand side of Eq. 1) depends on the job f being scheduled. Consider the expression C # # l b, s , f # # C # # l 0, f , s , f # # , which is commonly known as stretch factor [3]; its denominator is known as the service time of job f at site s. The latter depends only upon f and s, but not on the b. Further, the completion time C # # l b, s , f # # can be expressed as a sum of service time and waiting time: 2) C # # l b, s , f # # = C # # l 0, s ....

D. Ferrari, G. Serazzi, and A. Zeigner, Measurement and Tuning of Computer Systems, Prentice-Hall, Englewood Cliffs, NJ, 1983.

....Chitra94 also flags trace data as numerical, categorical, or mixed, so that the analyst can be warned if a numerical analysis method (e.g. visualization using a Gantt chart or autocorrelation) is applied to categorical data. 1. 4 Related Work Little of the workload modeling literature (e.g. [16, 5, 9]) is concerned with representing the dynamic or time dependent behavior of workloads. In recent years visualization tools have provided a new way to explore dynamic behavior in systems. General purpose commercial scientific visualization tools, such as PV WAVE and AVS, can be used. In addition ....

D. Ferrari, G. Serazzi, and A. Zeigner. Measurement and Tuning of Computer Systems. Prentice-Hall, London, 1983.

....seen as a combination of two main components that operate at different levels of the system and collect performance data using different techniques. This division in based on the interaction between the monitor and system, the technique of instrumentation used and the nature of the data collected [9]. The Kernel Module runs independently of the Web server and collects information about the operating system as a whole. The code of the Server Module is actually linked with the server code, and therefore runs as part of the server. It collects information about server performance during the ....

G. Serazzi D. Ferrari and A. Zeigner. Measurement and Tuning of Computer Systems. Prentice Hall, Englewood Cliffs, 1983.

No context found.

D. Ferrari, G. Serazzi, and A. Zeigner. Measurement and Tuning of Computer Systems.Prentice-Hall, 1983.

No context found.

D. Ferrari, G. Serazzi, and A. Zeigner. Measurement and Tuning of Computer Systems. Prentice-Hall, 1983.

No context found.

D. Ferrari, G. Serazzi, and A. Zeigner. Measurement and Tuning of Computer Systems. Prentice-Hall, Inc., 1983.

No context found.

D. Ferrari, G. Serazzi, and A. Zeigner. Measurement and Tuning of Computer Systems. Prentice Hall, Inc., Englewood Cli#s, 1983.

No context found.

D. Ferrari, G. Serazzi and A. Zeigner, Measurement and Tuning of Computer Systems, Prentice-Hall, Englewood Cliffs, 1983.

No context found.

D. FERRARI, G. SERAZZI, and A. ZEIGNER, Measurement and Tuning of Computer Systems, Prentice-Hall, 1983.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC