This action might not be possible to undo. Are you sure you want to continue?

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

13(2):23-33, 2003 ©

**Optimizing Computer System Configurations
**

Jozo J. Dujmović

Abstract - We present a quantitative method for selecting optimum configurations of computer systems. For each configuration we specify a set of requirements reflecting user’s needs. The level of satisfaction of requirements is called the global preference score. Using this indicator and the total cost of each configuration we solve the following optimization problems: (1) select a configuration which maximizes the global preference score for a constrained cost of computer system, (2) select a configuration which attains a given level of global preference for a minimum cost, and (3) select a configuration which attains the maximum global preference/cost ratio. We first present the optimization method and then a complete case study of computer optimization.

Decision makers sometimes reduce the number of possible configurations by selecting some fixed parameters of analyzed systems. However, the number of available configurations generally remains large, and it is useful to have optimization algorithms that can find the optimum configuration without exhaustive search of all possible combinations. Table 1. Configuration options of a desktop system No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Hardware Component Processor (speed) Memory size Monitor Video Card Hard Drive Modem CD ROM Drive Keyboard Network Card Tape drive Sound Card Speakers Mouse Floppy/Zip Drive Printer Number of Options 2 2 4 3 4 2 2 2 3 2 2 4 1 2 3 442,368

1. Introduction

Evaluation and selection of computer systems is a problem of selecting the most appropriate configuration of hardware and software for a specific computer user. This is not a simple problem because the majority of computer systems can have a large number of different configurations. Even for small and inexpensive computers it is possible to create thousands of different configurations. Each of these configurations has a specific cost and a specific degree of satisfying user requirements. In this paper we address a practical problem of finding the optimum configuration for a given user. Our method is primarily oriented toward optimizing large, complex, and/or expensive systems. However, to simplify the presentation and show a complete case study, we will use a relatively simple example of optimizing a desktop system. To illustrate the problem of large number of possible configurations Table 1 presents hardware options of a legacy desktop system [1]. Theoretically, there are 442,368 different hardware configurations in the price range from $2,022 to $4,867. The presented desktop system has a single processor, single disk unit, and standard input/output units for a single user. On the other hand, today’s servers often have more processor and memory options, multiple disk arrays, archive tapes, printers, and large and adjustable number of client systems. This substantially increases the number of possible configurations. The configuration selection process can be further complicated by other options related to software, service, and accessories. For example, if the presented desktop system comes with two levels of service, accessories, and optional software products then the total number of possible combinations becomes greater than 50 million.

The original manuscript received January 26, 2003. The revised manuscript received July 15, 2003. Jozo J. Dujmović, Department of Computer Science San Francisco State University

Total Number of Hardware Configurations

In this paper we propose an efficient method for optimizing computer configurations, based on the LSP method for system evaluation [3,4]. The proposed optimization method is general and can be applied to all advanced computer and communication systems (servers, workstations, mainframe systems, supercomputers, networks, etc.). Our presentation includes the proposed system evaluation and optimization methodology, followed by a characteristic case study.

n is called the cost function. n . 1 we have n = 15 and X 2 and C2 denote respectively the capacity and the cost of main memory. 64.. (4) floppy disks. then X 2 ∈ {32.. 400. (3) tape. or (2) to minimize the total cost while the total value is greater than a given threshold value. 800} .. i = 1. System Optimization Method The selection of an optimum computer configuration depends on two main parameters: the total cost and the total value of each analyzed configuration.2 Configuration options and elementary criteria The second step in the evaluation process is to create elementary criteria for evaluation of performance variables. Memory External M. i = 1. and the corresponding optimum values of performance variables X iopt = hi (Ciopt )... Data Com. 96. i = 1. Modem Network Ctrl Input Units Main Mem. I/O units consist of input units and output units. (7) modem. and the 6.. We will use the LSP method (which is one of cardinal ranking methods [5]) to compute the indicator of total value called the global preference. . Peripheral units consist of communication units and I/O units.4 GB disk can be considered to satisfy all applications (i. The decomposition process eventually yields the list of components whose parameters should be optimized: (1) processor. If the memory comes in 32 MB increments. 2.3GB disk in 80% of applications. 2.. (12) speakers. 600.. Such components are called performance variables.2GB. The global preference can be interpreted as the percentage of fulfilled user requirements for the evaluated system as a whole.. This determines the optimum configuration of the analyzed system according to a selected optimization criterion. i = 1... The preference is denoted Ei and we assume 0 ≤ Ei ≤ 100%. 2. The goal of the optimization process is to determine the optimum allocation of costs Ciopt . OPTIMIZING COMPUTER SYSTEM CONFIGURATIONS 2. if disk capacity may be 2.. which is used to compute the elementary preference for a given value of performance variable. n . The analyzed hardware consists of central units (processor and memory) and peripheral units.. (6) hard disk. (13) printer. Subsequent steps include the optimization process. X n . 4. (14) video card. 3. In the area of software acquisition this approach is known as the bestvalue technique [2]. or (3) to maximize the value/cost ratio. i = 1. These components are leaves of the decomposition tree: they cannot be further decomposed.1GB disk can be considered to properly satisfy only 20% of needs. the 4. We will now separately present each of the above steps. in Fig.. and 6.4 GB. Disk Video Ctrl Monitor Figure 1.2GB disk can perform satisfactorily in 50% of applications. The goal of optimization can be (1) to maximize the total value for a constrained cost of the system. J.. In all three cases the optimization process includes the following steps: Processor HARDWARE UNITS Processor and memory I/O and Data Com. I/O Units Keyboard Mouse Tape Unit Floppy Disk Output Units Disks Hard Disks Audio Units Sound Ctrl CD-ROM Hard Copy & Video Units Video Output Speakers Printer Mag. and (15) monitor. 5.. Each value of performance variable is related to a specific cost: X i = hi (Ci ). (11) sound card.3GB.. In the case of system optimization we select only those components that affect the cost of the system.1 Binary tree of performance variables The first step is to develop a list of all relevant components for evaluation and optimization of the analyzed system. (8) network card..e. For a specific user the 2. We recommend the use of binary trees because they substantially simplify the optimization process (Section 2. 100%).4). etc. (9) keyboard. and each increment costs $200. A binary tree for decomposition of basic hardware components 1.. Performance variables are denoted X1.. For example. For example. n . is called the elementary criterion. The first three steps are necessary to create a quantitative model for computing the global preference score of a computer system according to the LSP method. The function Ei = Gi ( X i ) . The percent of satisfaction of user's requirements is called the preference.. 1. n . (5) CD-ROM. The corresponding inverse function Ci = hi−1( X i ).1GB. then each capacity satisfies a given fraction of user's needs. The list of performance variables can be derived using various decomposition trees.. (10) mouse. the 3. (2) internal memory..24 DUJMOVIĆ. 3..128} and C2 ∈ {200. Binary tree of performance variables Configuration options and elementary criteria Preference aggregation structure Optimum preference and cost allocation functions Computation of optimum configuration Each of these components has a cost and for each of them we can define a single performance variable. An example of the binary tree for hardware units of a desktop system is shown in Fig. 4.

Table 2 illustrates the elementary criterion and the cost function for the capacity of a hard disk drive.. the elementary criteria yieldW W The global preference E0 denotes the global level of satisfying user's requirements. To aggregate the first two elementary preferences E1 and E2 we use the following weighted power mean: r r 1/ r E12 = (W1E1 + W 2 E2 ) . The final step in a system evaluation process is to aggregate all elementary preference scores and compute the global preference for the whole system: E0 = L(E1. E 2 ) .3 49 109 50 80 alized using aggregation operators based on weighted power means. In other words... The aggregation process generates the global preference as an indicator of the value of the whole system.. The weights W1 and W2 express the relative importance of the first and second input. E1 W1 E12 E2 E3 E4 r W12 W2 W3 W 34 E1234 s E34 W4 4 6. here we will present only the basic ideas. In both cases if either E1 = 0 .. [$] Preference [%] 1 2. Using these values and the corresponding elementary criteria we can compute n elementary preference scores: E1 . and M490. and cost functions for all components of the analyzed system. the cost row denotes only the cost difference between the basic cost of the 2. For some components it is difficult to define a measurable performance indicator. or with three types (models) of speakers. En ) = L(G1( X 1)..3 Preference Aggregation Structure A specific configuration of a computer system is defined by the values of n performance variables X 1. X n ). In this case the configuration always includes a hard disk. or E 2 = 0 . Consequently. The elementary preferences reflect the value of individual components... Table 2 Hard Disk Drive Capacity [GB] Cost Dif. The intensity of this operator can be selected using appropriate values of r . − ∞ ≤ r ≤ +∞ . The same function is used in all aggregation nodes. 2.. This case is exemplified in Table 3. the consequence is E12 = 0 ..4 184 100 q Figure 2. For example. The selection of parameters can be done according to methods described in [7]... M290. denoted M90. Table 3 Type Cost [$] Preference [%] 0 0 0 0 1 M90 49 70 Option 2 M290 124 90 W1 + W2 = 1.2 4. E12 = 1 / (W 1 / E1 + W 2 / E 2 ) .. but the purpose of evaluation is to determine the value of a system as a whole. and the weighted power mean becomes the geometric mean E12 = E1 1 E 2 2 . elementary criteria (evaluation functions).JOURNAL OF AUTOMATIC CONTROL.. The model number can be used as the performance variable..1 0 20 Option 2 3 3. The logical properties of the aggregation function depend on the selected value of parameter r. and the cost of the smallest disk (option 1) is included in the basic cost of the system. For example if r = −∞ the weighted power mean becomes conjunction. The analyzed computer can be configured without speakers (option 0).. A simple binary preference aggregation tree Suppose we aggregate elementary preferences using a binary tree exemplified in Fig. E n . 2.1 GB disk and the cost of higher capacity disks. If the user needs simultaneous satisfaction of the first and second elementary criterion then it is necessary to use values r < 1 and the weighted power mean becomes a simultaneity operator...Gn ( X n )) = G( X 1. 0 ≤ E0 ≤ 100% . The aggregation function L is re- . More details about the aggregation functions can be found in [3] and [4]. which requires even more simultaneity between E1 and E 2 than the geometric mean. The exponent r is used to adjust the logical properties of this function (simultaneity or replaceability of inputs). The properties of this function depend on its parameters (weights and exponent). and for each model we can define the corresponding preference and cost. UNIVERSITY OF BELGRADE 25 The simplest notation of elementary criteria and the corresponding cost functions is a tabular form. which is the highest intensity of the simultaneity requirement. and each time the parameters are adjusted to reflect a desired evaluation strategy. X n . A relatively low simultaneity requirement is obtained if r = 0 . For r = −1 we generate the harmonic mean.. E12 = min(E1. Speakers 3 M490 189 100 At the end of this step we have configuration options..

X n ) . OPTIMIZING COMPUTER SYSTEM CONFIGURATIONS ing E1 and E 2 must be simultaneously satisfied. if E 1 > 0 and E 2 = 0 this yields E 1 > E 12 > 0 . The partial absorption operator combines a mandatory requirement (input E 1 ) and a desired feature (input E 2 ).8 9.1 −∞ If r > 1 the weighted power mean can be used to model replaceability of inputs. If E 2 > E 1 > 0 then E 2 > E 12 > E1 > 0 .7 -3.. Partial absorption function A very useful and frequently used compound aggregation operator is the partial absorption. E0 = H (C1. n .52 20. The operator A denotes the arithmetic mean and C denotes one of simultaneity operators (used as an AND gate).. Consequently. n . and the simultaneity operator behaves similarly to an AND gate... 3... i = 1. Using suitably selected aggregation functions we can realize a preference aggregation tree that exactly reflects the needs of a specific user. The basic optimization problem consists of finding the maximum value of E0 for C0 ≤ C . Eight increasing levels of simultaneity used in most practical cases (see [4]) are denoted using symbols C − − (lowest simultaneity level.4 Optimum preference and cost allocation functions Each value of performance variable X i is a function of cost Ci . . Figure 3. J. E 1 is a Let C denote the available budget for computer acquisition..79 3.93 5. opt opt C1 +.26 DUJMOVIĆ. A solution of this problem for a variable total system cost C is a set of optimum allocation functions Ciopt = ai (C ). which yields the optimum configuration paC1 . r=0...62). Obviously... r = −∞ ) as follows: Simultaneity operators and the corresponding values of r C-CC-+ CA C+C+ C++ C 0. A lower replaceability level can be obtained for r = 2 when we 2 1/ 2 2 + W2E2 ) ......+Cn . E12 = max(E1. The optimization problems can now be formulated as follows: • • • Find the optimum system configuration yielding max the maximum global preference E0 for a given constrained cost of the system C0 ≤ C (C is the maximum acceptable value)... Once we know the global preference function we can develop optimization algorithms to provide a suitable combination of high global preference and low global cost.. pure conjunction. E 12 = W1E1 + W 2 E 2 .45 2. The case r = 1 yields the traditional weighted arithmetic mean. the global preference of a system can be expressed as a function of individual costs.62 0.. primary input while E 2 is much less important and used to slightly increase/decrease the output preference E 12 with respect to the primary input E 1 . E 2 ) . Cn ) ..15 -.+Cn i = 1. In this case the weighted power mean becomes disjunction. n ≤C which determine the optimum allocation of the total cost C and the parameters of the optimum configuration: X iopt = hi (Ciopt ) = hi (ai (C )). If the mandatory requirement is not satisfied ( E 1 = 0 ) then E 12 = 0 regardless the value of E 2 .g.26 -.02 2. the maximum value of the ratio Q = E0 / C0 ) E1 E2 mandatory desired 1 − W1 W1 W2 1 − W2 The solution of the above problems can always be expressed as an optimum distribution of individual costs opt opt . The highest replaceability level is obtained if r = +∞ . i = 1. min Find the minimum cost of a system C0 necessary to attain a given level of global preference E0 ≥ E (E is the minimum acceptable value).Cn A C E12 rameters X iopt = hi (Ciopt )... Find the best value configuration which attains the maximum preference per unit of cost (e. On the other hand. and the global cost is C0 = C1 +.72 -1. and it is frequently used either alone or combined with other operators..5 -9.. shown in Fig. A global quantitative indicator of satisfaction of user needs is the global preference E0 = G( X 1.6 +∞ In all cases the replaceability operator behaves similarly as an OR gate. and it is sufficient to satisfy only one of the input criteria.. to D (highest replaceability) similarly as in the case of simultaneity: Replaceability operators and the corresponding values of r D-DD-+ DA D+D+ D++ D 1. Eight have the quadratic mean E12 = (W1E1 practical levels of replaceability are denoted using symbols D − − (lowest replaceability level). to C (highest simultaneity... 2. This case is neutral (in the middle between simultaneity and replaceability).

C2 ) opt = E12 (C1 (C12 ).. 2. max E12 (C1.C2 ) = (W1E1 At the end of this process we get the function opt C0 a E 0 (C0 ) that shows the maximum preference that r 1/ r = [W1(G1( X 1)) + W2 (G2 ( X 2 )) ] r can be achieved for a given global cost C0 ...C2 (C12 )) Figure 4. and compute the minimum cost and the maximum cost of the node: min min min max max max C12 = C1 + C2 . For example. This process is shown in Fig. 5 and can be summarized as follows: .JOURNAL OF AUTOMATIC CONTROL. 6. The total allocated resources in this node are C12 = C1 + C2 .. Consequently. and C2 (C12 ) as four arrays of values: C12[.].C2 (C12 )). and C2opt[. E12opt[. use opt opt opt E12 (C12 ) and E 34 (C34 ) to compute E1234 (C1234 ) min min max max in the range C12 + C34 ≤ C1234 ≤ C12 + C34 .5 Computation of optimum configurations = [W1(G1(H1(C1)))r + W2 (G2 (H 2 (C2 ))) r ]1/ r This simple linear search yields the optimum cost values opt C1 (C12 ) and C2 (C12 ) which give the maximum value opt of the subsystem preference E12 (C12 ) for the subsystem cost C12 .. We also have a sequence of optimum cost allocation functions. The same process can be repeated for all values min max of cost in the range C12 ≤ C12 ≤ C12 .. According to Fig.C2 ) be the aggregate preference of inputs 1 and 2 as a function of costs of the first two inputs.. 5 this step is denoted as “allocate”. ≥ H (C1.].C2 ) max E 12 (C12 ) Select a layer of aggregation nodes. C1 (C12 ) .]. Computation of the optimum preference function 4.. The range min C2 • The optimum cost allocation functions: opt opt C1 (C12 ) and C2 (C12 ) opt opt opt Save functions E12 (C12 ) . The optimum allocation functions can be found separately for each node in the binary preference aggregation tree. we have to search for the optimum allocation of resources in the range min min min max max max C1 + C2 = C12 ≤ C12 ≤ C12 = C1 + C2 . e. as well as the optimum cost allocation functions opt opt C12 (C1234 ) and C34 (C1234 ) . Let E12 (C1. UNIVERSITY OF BELGRADE 27 For each cost distribution satisfying C1 +.C2 ) opt C12 max C12 = E12 (C1 (C12 ). Cn ) . Figure 4 illustrates the optimization procedure for the first two parameters that have costs C1 and C2 . r r 1/ r + W2 E2 ) E12 (C1. 2.. In the selected layer select a node..+Cn ≤ C we have opt opt H (C1 .g. In Fig.. starting with the first layer. e... C1 + C2 = C12 C1 m in C 12 o pt C1 (C 1 2 ) E12 (C12 ) = opt C1 + C 2 ≤ C12 opt max E12 (C1. C1opt[. opt The final step in the optimization process is to compute the parameters of optimum configurations. opt E12 (C12 ) E12 C2 C12 opt C2 (C12 ) E12 (C1.. Cn ) 1. In the second and higher layers use as inputs the previous optimum preference functions.. The system optimization procedure consists of repeating the presented node optimization operations for all nodes in the binary aggregation tree. the first node 12. These arrays can be later used to compute the values of inverse functions. of these costs is min max C1 ≤ C1 ≤ C1 and . for each value of C12 we can locate the maximum preference along the line C1 + C2 = C12 : E12 (C12 ) = opt C1 + C 2 = C12 opt max ≤ C2 ≤ C 2 5.]... 4. C1 and C2 . C12 = C1 + C2 min max For all node costs in the range C12 ≤ C12 ≤ C12 compute the optimum preference function and two optimum allocation functions: • The optimum preference function: 3. opt opt E12 a C12 (E12 ) Repeat steps 2 to 4 for all nodes in the selected layer Repeat steps 2 to 5 for all layers in the binary tree.g..

C1234 . This case is also illustrated in Fig. The second optimization problem is to find the configuration that attains a given global preference for a minimum cost. In this case. 5.28 DUJMOVIĆ. Computation of optimum preferences and computation of optimum configurations The computation of optimum configurations is similar for all three basic optimization problems. Using the desired global preference E1234 and the optimum prefopt erence function C1234 a E1234 (C1234 ) (recorded as arrays . Finally. using the example in Fig. i.e. OPTIMIZING COMPUTER SYSTEM CONFIGURATIONS C1 E1(C1) 12 OPT C12 opt C1 (C12 ) opt E12 (C12 ) opt C2 (C12 ) opt C12 (C1234 ) opt opt E1234 ) 1234(C1234 1234 ) opt C34 (C1234 ) C2 E2 (C2 ) C1234 opt E1234 1234 OPT C3 E3 (C3 ) 34 OPT C4 E4 (C4 ) C34 opt C3 (C34 ) opt E 34 (C34 ) opt C4 (C34 ) OPTIMIZE ALLOCATE opt C1 opt X1 h1(C opt 1 ) opt C1 (C12 ) opt X2 opt h2 (C2 ) opt C2 E (C12 ) opt C2 (C12 ) opt 12 opt C12 opt C12 (C1234 ) opt X3 opt h3 (C3 ) C opt 3 E1234 opt C3 (C34 ) opt E34 (C34 ) opt E1234 (C1234 ) opt C34 (C1234 ) opt C34 C1234 X opt 4 opt h4 (C4 ) opt C4 opt C4 (C34 ) Figure 5. we first use the desired global cost. This technique can be applied to find all other optimum allocation functions and configuration parameters. 5. The first problem is to find the configuration which attains the maximum global preference for a given global cost. X2 = h2 (C2 and Since = opt opt (C12 (C1234 ))) h1(C1 we have that the first opti- opt opt mum allocation function is a1(C1234 ) = C1 (C1 2 (C1234 )) . J. we use C1 and C1 ) and C2 opt C2 to compute the optimum configuration parameters opt opt ) X1 = h1(C1 opt X1 opt opt ). to compute the optimum allocation in nodes 12 and 34. We then continue in the opt to determine same way: we go to node 12 and use C12 opt opt opt opt opt (C12 (C12 ) . opt opt C12 (C1234 ) and C34 (C1234 ) .

The criterion is specified by estimating the level of satisfaction . the speakers and printer can be optional in corporate settings) we explicitly list the 0 option.51%.253 is required to satisfy 90% of user needs. A demo version of OCCS is available on the Internet [6].1GB) is specified as $0 because it is included in the base package’s price ($2... Therefore. A case study of computer optimization In this section we present a complete example of optimization of a legacy desktop system [1]. In our example. This search yields the cost C1234 that corresponds to the optimum configuration.g. This method is very efficient and can be applied for any practical value of n . The result of preference aggregation is the global preference for the system as a whole. and other costs are incremental with respect to option 0.1GB to 6. while disjunctive operators (DA) specify replaceability. The optimum-preference-over-cost curve depicts the level of satisfaction of user needs per monetary unit.. 8. They are expressed as percentages. For systems with n input parameters the number of aggregation nodes is n − 1 . The resulting curve shows all configurations that achieve the maximum preference for a given value of global cost. For components that are not included in the base package (e. The final optimization results are presented in Figure 7. Each component of the COT includes the full definition of the corresponding elementary criterion. All preference values in Table 4 are selected as realistic examples. A user capable of precisely expressing the needs for all components of the system can use the presented LSP approach to maximize the return on computer investment. Three important configurations are presented in Table 5. Assuming that a mouse and a CD-ROM are included in all configurations. The costs of other disks represent the differences between their full cost and the full cost of the entry-level disk. Optimum configurations in the range from $2. Therefore. OCCS (Optimum Configurator of Computer Systems).]) we can determine the corresponding cost C1234 .022 specified in the processor line). Conjunctive operators (C-+. full prices are listed in COT. UNIVERSITY OF BELGRADE 29 C1234[. in the case of the EIDE Hard Disk. and the corresponding costs and preferences. The optimization process generates parameters of optimum configurations. and (3) the least expensive system satisfying 90% of requirements. and the sum of input weights for each block is 100%. The option 0 denotes the absence of the analyzed component. the maximum level of satisfaction achievable for under $2. Table 4 illustrates the case where 32MB of main memory satisfies 75% of user’s requirements. These elementary criteria are then used for computing elementary preferences for all values of performance variables.] and E1234opt[. For example. Similarly. We will use the tree shown in Fig. the computation time for the whole optimization process is of the order O[ n − 1] . (2) the best value configuration.4GB). the corresponding structure is shown in Figure 6. 3.g. 4]. Another result of optimization is shown in Fig. CA. To help in this direction we have developed a specialized tool.500 achieve high levels of preference over cost. Using this cost the problem is reduced to the previous case of finding the optimum configuration for a given global cost. The third step in the optimization procedure is the development of the preference aggregation structure for computing the global preference. which yields 15 performance variables.500. $3. with each option. For example. In cases where this option is acceptable (e. Each aggregation block includes weights that denote the relative importance of inputs. This type of results is crucial for management decision making and planing. The presented optimization procedure solves all three basic system optimization problems.12% of user requirements. printers).500 is 70. They are (1) the best configuration under $2. The highest point defines the best-value configuration that carries a price tag of $2.JOURNAL OF AUTOMATIC CONTROL. In general.651 and satisfies 76. The cost of the entry-level disk (2. and 64MB is sufficient to satisfy all requirements. Aggregation operators specify desired logic relationships between input preferences [3. option 1 cost is $0 for components that are part of the base package. and the problem is again reduced to the previous case.500 to $3. the adjustable parameter is the capacity (from 2. 1. The third optimization problem is to find the configuration that attains the maximum preference/cost ratio. yielding optimum configuration parameters. and C+) specify the simultaneity of inputs. For each performance variable we specify the adjustable values (options). and can be adjusted to reflect other types of users. The second step is to create the Configuration Option Table (COT). Using typical data from [1] the corresponding COT includes 14 adjustable parameters (performance variables) shown in Table 4. this criterion reflects the specific needs of the selected class of users. The first step of the optimization procedure is to develop a binary tree of performance variables.. The optimization algorithm then compares the cost-preference pairs. Using this model we compute the global preference and the corresponding total cost for each configuration. It consists of a sequence of optimizations of individual nodes in the binary aggregation tree. opt (C1234 ) / C1234 can The maximum value of the ratio E1234 be easily found using the linear search of the array E1234opt[i]/C1234[i].

6Kbd 129 100 12-16X 0 80 Spacesaver 0 50 3C900 PCI 99 85 4GB 199 100 WaveSyn SW 0 70 ACS 90 49 70 1.7”T 0 40 Virge 2MB 0 30 2.9”T 340 85 Maxtor 4MB 139 100 4. J.9” 119 60 Virge 4MB 40 60 3.1 GB 0 20 33. OPTIMIZING COMPUTER SYSTEM CONFIGURATIONS Table 4.2 GB 49 50 3 4 10HS 15.3 GB 109 80 20TD 19”T 939 100 6.44+100MB 138 100 HP LJ 6L 409 100 ACS 490 189 100 .30 DUJMOVIĆ. Elementary Criteria Configuration Options and Elementary Criteria Processor Speed Cost Preference Main Capacity Memory Cost Preference Monitor Type Cost Preference Video Board Type Cost Preference Hard Disk Capacity Drive Cost EIDE Preference Modem Speed Cost Preference CD-ROM Speed Drive Cost Preference Keyboard Type Cost Preference Network Type Card Cost Preference Tape Capacity Drive Cost Preference Sound Type Card Cost Preference Speakers Type Cost Preference Floppy Capacity Disk Cost Preference Printer Type Cost Preference Option 0 1 200 MHz 2022 85 32MB 0 75 8HS 13.44MB 0 75 HP DJ 820 309 65 2 233MHz 2100 100 64MB 239 100 10LS 15.4 GB 184 100 0 0 0 0 0 0 0 0 0 12 / 24X 25 100 Performance 29 100 3C905-TX 109 100 0 0 0 0 0 0 WaveTbl 32 39 100 ACS 290 124 90 1.

Preference aggregation structure for computing the global preference 100% 90% 80% 70% Preference 60% 50% 40% 30% 20% 10% 0% $0 $500 $1.000 $3.500 $2.000 $2.500 Cost $3.000 $1.JOURNAL OF AUTOMATIC CONTROL.000 $4. UNIVERSITY OF BELGRADE 31 Processor speed Internal Memory Hard Disks CD-ROM Drive Floppy Disks Archive Tape Monitor Video Controller Printer Sound Card Speakers Keyboard/Mouse Network Card Modem 70 30 DA Communications 60 40 CA Audio Output 20 70 30 C-+ 27 73 A 37 63 C-+ Visual Output 80 A 50 50 CA 90 10 I /O Units Peripherals C-+ 70 CA 30 35 65 C+ 89 11 C-+ 90 10 C-+ 45 55 A 26 74 C-+ 45 55 CA 80 20 CA Central Subsystem External Memory PC Figure 6.500 $5.500 $4. Maximum preference for constrained cost of a desktop system .000 Figure 7.

4 GB 12/24X 1.05 0.500 $4.000 $1.00 $0 $500 $1.477 70.000 Figure 8.32 DUJMOVIĆ. Optimum Configurations Configuration Parameter Processor speed Memory capacity Disk Capacity CD-ROM Speed Floppy Disks Tape Drive Monitor Video Card Printer Sound Card Speakers Keyboard Network Card Modem Total Cost Global Preference Best configuration under $2.44 MB None 1000LS 15.500 $2. OPTIMIZING COMPUTER SYSTEM CONFIGURATIONS 0.20 0. J.9” Maxtor 4MB None WaveSynth SW ASC90 Performance 3C905-TX None $2.000 0.9” Trinitron Maxtor 4MB None Wavetable AWE32 ASC90 Performance 3C905-TX None $3.000 $3.500 $5.10 0.51% Best value configuration (maximum Preference/Cost) 200 MHz 32 MB 6.500 Cost $3.44 MB None 1000LS 15.64% .44 MB None 1000HS 15.253 89.15 0.30 0.651 76.12% Least expensive system satisfying 90% of requirements 233 MHz 64 MB 6.500 200 MHz 32 MB 4.000 $2.3 GB 12-16X 1.9” Virge 4MB None WaveSynth SW ASC90 Performance 3C905-TX None $2.4 GB 12-16X 1.000 $4. Maximum preference over cost Table 5.25 Preference per $1.

In addition to optimum configurations the proposed method also provides background information necessary for justifying the proposed decisions.. IASTED/ACTA Press. 5. These components can be specified either through the values of their parameters. Making Multi-ple-Objective Decisions. or through other non-numerical descriptors. as well as with volume purchases of smaller computers. 1998. M. It is possible to optimize all system components that contribute to the total cost.0. edited by M. Proc.com. Our models include adjustable parameters in order to express a wide spectrum of realistic user requirements. Our optimization method is general. and J. 1994. 1993. 19-43. ISBN: 0-88986-244-3. accessories. pp. Vice versa. The proposed optimization method is very fast: the run time is directly proportional to the number of components of the optimized system. for each desired level of user requirements we can find the minimum necessary cost. J. communication equipment. of the 32nd SICE Annual Conference.J. software. The 22nd International Conference for the Resource Manage-ment and Performance Evaluation of Enterprise Com-puting Systems. pp. Dujmovic. substantial reduction of cost can be achieved by the combined optimization of configuration and insignificant reduction of user satisfaction. Interfaces. the use of optimization techniques presented in this paper is fully justified and can result in substantial savings. pp. A Method for Evaluation and Selection of Complex Hardware and Software Systems. The implementation of the method is not complex and we have developed a corresponding optimization tool whose demo version is available on the Internet [6]. No. 1997. IEEE Computer Society Press. T. This is the case with procurements of medium and large systems and networks. 1. 13-46 . For each level of cost the proposed method determines the corresponding maximum level of satisfying user requirements. pp. Vol. 1.dell. Saaty. Quantitative Evaluation of Software. Within this range.seas.JOURNAL OF AUTOMATIC CONTROL. pp. 24. Conclusions All substantial investments in computer equipment require fine tuning of computer configurations. [3] [4] [5] [6] [7] . Pet-Edwards. 6..J. In such instances. References [1] [2] Dell Computer Corporation. Dujmovic. UNIVERSITY OF BELGRADE 33 4. We precisely identify the components of all optimum configurations. 1997. Version 2.H. etc. Mollaghasemi. Vol. 1996. the configurations in the upper half of the price range. 368-378. CMG 96 Proceedings. Optimum Configurator of Computer Systems (OCCS). http://www.1379-84. 3-7..com .L. and can be applied for optimum selections of a variety of computer systems: hardware. Department of the Air Force. Proceedings of the IASTED International Conference Software Engineering. June 1996. Hamza. Dell Dimension XPS. http://www. Guidelines for Successful Acquisition and Management of Software-Intensive Systems. Vol. J. In our case study of configuration optimization. 1997.13-55. span less than 10% of user needs. How to Make a Decision: The Analytic Hi-erarchy Process.

- Ex Instalacije
- Zakon o Eksplozivnim Materijama
- RAD Pointing Mechanism for Optical Communication
- AN9420 Strujni Pojačavač Teorija
- export regulations
- Us6211951 Boresight Alignment Method
- Tutorials on Light Measurement
- Kb Dynamic Range 20100813
- kb_smear_20100723
- kb_smear_20100723
- kb_hot_pixel_20100813
- Urban Operations
- FPGA to ASIC Conversion Reference Manual
- LC898214XC - Auto Focus (AF) Controller
- vojno delo, 2013
- Innovations Enable Streamlined Monitoring
- Post-Cold War Destinies of the Russian, Ukrainian, Polish and Belarusian Defence Industries
- Computational Social Science and System Dynamics
- RAD Holst Imaging System Fundamentals
- Methods for NVG Visual Acuity Determination 3
- Methods for NVG Visual Acuity Determination 2
- Methods for NVG Visual Acuity Determination
- Laser Aimer
- Reflex Sighting System
- Night Vision Monocular

quantitative method for selecting
optimum configurations of computer systems

quantitative method for selecting

optimum configurations of computer systems

optimum configurations of computer systems

- Branch & bound: An Algorithm for Divisible Load Scheduling with Result Collection on Heterogeneous Systemsby International Journal of Science and Engineering Investigations

- IJERD (www.ijerd.com) International Journal of Engineering Research and Development IJERD : hard copy of journal, Call for Papers 2012, publishing of journal, journal of science and technology, research paper publishing, where to publish research paper, journal publishing, how to publish research paper, Call For research paper, international journal, publishing a paper, how to get a research paper published, publishing a paper, publishing of research paper, reserach and review articles, IJERD Journal, How to publish your research paper, publish research paper, open access engineering journal, Engineering journal, Mathemetics journal, Physics journal, Chemistry journal, Computer Engineering, how to submit your paper, peer review journal, indexed journal, reserach and review articles, engineering journal, www.ijerd.com, research journals, yahoo journals, bing journals, International Journal of Engineering Research and Development, google journals, journal of engineering, online Submisby IJERD

- Controller Area Network (CAN) Schedulability Analysis With FIFO Queues
- ngs06
- Article 05
- Particle swarm optimization Relative Paper
- Graph Matching Algorithm for Task Assignment Problem
- Branch & bound
- Classifier MCT for immediate mode independent task scheduling in Computational Grid
- A Tutorial Introduction to Research on Analog and Mixed-Signal Circuit Testing
- Wave Let
- Encode
- 00857086
- Robust Calibration for Localization in Clustered Wireless Sensor Networks-0Rm
- Performance of Algorithms for Scheduling
- Social Network GPU
- Pre Port
- Analyzing the Decoding Performance of Rate ⅓ Convolutional Code
- A New Method for Subgraph Detection - SubGraD
- 03A.2
- 01250830
- PCG Analysis IEEE
- IJERD (www.ijerd.com) International Journal of Engineering Research and Development IJERD
- Comparative Study of Ant Colony Optimization And Gang Scheduling
- Hwang 2010
- A Fixed-Die Floor Planning Algorithm Using an Analytical Approach
- FPGA Implementations of High-Dimensional Cube Testers on the Stream Cipher Grain-128
- International Journal of Engineering Research and Development (IJERD)
- A Study of Load Distribution Algorithms in Distributed Scheduling
- Timing and Code Size Optimization on Achieving Full Parallelism in Uniform Nested Loops
- Cf 33497503
- 2006_conference.pdf

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue listening from where you left off, or restart the preview.

scribd