## Time and cost tradeoff for distributed data processing (1989)

Venue: | Computers ind. Engng |

Citations: | 3 - 1 self |

### BibTeX

@ARTICLE{Lee89timeand,

author = {Heeseok Lee},

title = {Time and cost tradeoff for distributed data processing},

journal = {Computers ind. Engng},

year = {1989},

volume = {16},

pages = {553--558}

}

### OpenURL

### Abstract

Abstract--An important design issue in distributed data processing systems is to determine optimal data distribution. The problem requires a tradeoff between time and cost. For instance, quick response time conflicts with low cost. The paper addresses the data distribution problem in this conflicting environment. A formulation of the problem as a non-linear program is developed. An algorithm employing a simple search procedure is presented, which gives an optimal data distribution. An example is solved to illustrate the method.

### Citations

751 |
Nonlinear Programming: Theory and Algorithms
- Bazaraa, Sherali, et al.
- 1993
(Show Context)
Citation Context ...efine the auxiliary function X~(2) as follows: And also define The X(2) is monotonically increasing in 2, where X i .-~. -Y - -- -- 7 qJ,-dr' ! Vi = 1, . , . N. (17) # X,(2) = #--' 1 / x#i Vi = 1, N. =-=(18)-=- Y 7 X/ 2 - di N X(2) = ~ X~(2). (19) i=1 2 > max d~. (20) i6l The monotonicity of X(2) implies that we can easily calculate 2 numerically so that X~(2) = 1, thus implying that the constraint (1) is s... |

162 |
Comparative models of the file assignment problem
- Dowdy, Foster
- 1982
(Show Context)
Citation Context ...mmunication cost is minimized. Queueing models have focused mainly on minimal response time as a criterion of performance [9-11]. A comprehensive review of the file allocation problem can be found in =-=[12]-=-. Typically, one of two optimization objectives has been adopted in the design of distributed data processing systems: either minimization of overall operating cost or optimization of some performance... |

149 |
A proof of the queueing formula
- Little
(Show Context)
Citation Context ...verage time spent in the system by a file request. Denoting this random variable by T, we can see that T can be obtained as the weighted sum of response time of each node simply using Little's result =-=[14]-=-: T= L 7g L. (4) The g includes the request processing time, the delay times resulting from the processing of file request, and the computer access time. In most cases, the computer access time is muc... |

136 |
Distributed Databases. Principles and Systems
- Ceri, Pelagatti
- 1984
(Show Context)
Citation Context ...ertain advantages depending upon the manner in which data are distributed. The distribution of data can be fully characterized by two problems: designing the fragmentation and allocation of fragments =-=[1]-=-. Fragmentation design has rarely been studied. Some methodological approaches are presented in [2, 3]. On the other hand, the allocation of fragments has been widely analyzed in the context of the fi... |

65 | Optimal file allocation in a multiple computer system
- Chu
- 1969
(Show Context)
Citation Context ... been studied. Some methodological approaches are presented in [2, 3]. On the other hand, the allocation of fragments has been widely analyzed in the context of the file allocation problem, since Chu =-=[4]-=- originally investigated the file allocation on a multi-processor system. In [5-8], the objective of the problem is to allocate copies of data files to processing nodes so that the sum of data storage... |

36 |
Allocation of copies of files in an information network
- Casey
- 1973
(Show Context)
Citation Context ... Combined with the fact that the arrival process concerning ?~ in (3) is a Poisson process, each processing node can be modeled as M/M/I queue [15] in an open central server network so that Ti = --.l =-=(5)-=- #~- 7 x~ (1)sTime and cost tradeoff 555 To prevent the queue lengths from growing without bound, it is required that xi <l~, Vi= l ..... N (6) ? that is, the mean rate of requests transmitted to a pa... |

23 |
An Introduction to Database Systems, Vol I
- Date
- 1986
(Show Context)
Citation Context ...iew of the tradeoff between system response time and operating cost. SYSTEM ANALYSIS AND MODEL FORMULATION An information system in distributed data processing systems is almost invariably relational =-=[13]-=- due to its simplicity of fragmentation. In this case, the information system is a common, global relation itself. The relation is just a mathematical term for a table consisting of tuples and attribu... |

22 |
File placement in distributed computer systems
- Wah
- 1984
(Show Context)
Citation Context ... l ..... N (6) ? that is, the mean rate of requests transmitted to a particular node must be less than the mean file service rate. From (4) and (5), the system response time is N T= x, e= l/~i - ?xi" =-=(7)-=- We can express the operating cost in terms of the data distribution x~. The operating cost consists of storage cost and communication cost. The expected storage cost per unit time is N S = Y, six e. ... |

19 |
Optimal program and data locations in computer networks
- Morgan, Levin
- 1977
(Show Context)
Citation Context ...[15] in an open central server network so that Ti = --.l (5) #~- 7 x~ (1)sTime and cost tradeoff 555 To prevent the queue lengths from growing without bound, it is required that xi <l~, Vi= l ..... N =-=(6)-=- ? that is, the mean rate of requests transmitted to a particular node must be less than the mean file service rate. From (4) and (5), the system response time is N T= x, e= l/~i - ?xi" (7) We can exp... |

13 |
Distributed database design methodologies
- Ceri, Pernici, et al.
- 1987
(Show Context)
Citation Context ... can be fully characterized by two problems: designing the fragmentation and allocation of fragments [1]. Fragmentation design has rarely been studied. Some methodological approaches are presented in =-=[2, 3]-=-. On the other hand, the allocation of fragments has been widely analyzed in the context of the file allocation problem, since Chu [4] originally investigated the file allocation on a multi-processor ... |

13 |
Queuing systems: Computer applications
- Kleinrock
- 1976
(Show Context)
Citation Context ...equest is an exponential random variable with mean 1//~i. Combined with the fact that the arrival process concerning ?~ in (3) is a Poisson process, each processing node can be modeled as M/M/I queue =-=[15]-=- in an open central server network so that Ti = --.l (5) #~- 7 x~ (1)sTime and cost tradeoff 555 To prevent the queue lengths from growing without bound, it is required that xi <l~, Vi= l ..... N (6) ... |

11 |
Optimal file allocation in multilevel storage systems
- Chen
- 1973
(Show Context)
Citation Context ... is simply computed as Accordingly, the expected communication cost per unit time is Therefore the expected operating cost per unit time is or denoting se + ci by de, 1 N ce = ~j~= (~tjqj~ + fljuj,). =-=(9)-=- N a = ~ CiX i. (10) i=l N C = S + Q = ~, (s,+ G)Xe, (1 1) e=l N C = ~ dixe. (12) i=l Here de is the operating cost incurred at node i. From (7) and (12), it is pointed out that each of the two cost f... |

11 |
Nonlinear programming: A Unified Approach (Prentice-Hall
- Zangwill
- 1969
(Show Context)
Citation Context ... follows: And also define The X(2) is monotonically increasing in 2, where X i .-~. -Y - -- -- 7 qJ,-dr' ! Vi = 1, . , . N. (17) # X,(2) = #--' 1 / x#i Vi = 1, N. (18) Y 7 X/ 2 - di N X(2) = ~ X~(2). =-=(19)-=- i=1 2 > max d~. (20) i6l The monotonicity of X(2) implies that we can easily calculate 2 numerically so that X~(2) = 1, thus implying that the constraint (1) is satisfied. In this case, a search proc... |

7 |
Introduction to Numerical Computations
- Vandergraft
- 1983
(Show Context)
Citation Context ...notonicity of X(2) implies that we can easily calculate 2 numerically so that X~(2) = 1, thus implying that the constraint (1) is satisfied. In this case, a search procedure like the bisection method =-=[17]-=- can be used. The corresponding values of xi are then obtained from (17). The constraint (6) is always true of any values of x~ in (17). Hence the only way the set of x~ is not optimal is if one or mo... |

6 |
An integer programming model for the allocation of databases in a distributed computer system, European
- Pirkul
- 1986
(Show Context)
Citation Context ... We can express the operating cost in terms of the data distribution x~. The operating cost consists of storage cost and communication cost. The expected storage cost per unit time is N S = Y, six e. =-=(8)-=- /=1 Since the file request arriving at a node may result in updating a local or remote data file, the expected communication costs include query communication cost and update communication cost. Let ... |

3 |
A methodology for structured database decomposition
- Chang, Cheng
- 1980
(Show Context)
Citation Context ... can be fully characterized by two problems: designing the fragmentation and allocation of fragments [1]. Fragmentation design has rarely been studied. Some methodological approaches are presented in =-=[2, 3]-=-. On the other hand, the allocation of fragments has been widely analyzed in the context of the file allocation problem, since Chu [4] originally investigated the file allocation on a multi-processor ... |

1 | The file allocation problem--a queueing network optimization approach - Srinivasan, Kant - 1987 |

1 |
A simple method to solve simple allocation problems, liE Trans
- Vidal
- 1987
(Show Context)
Citation Context ..., Vi = 1 ..... N. (15)s556 HEESEOK LEE OPTIMAL DATA DISTRIBUTION The optimal solution to the problem (15) is obtained as follows. Our analysis is based on the well-known results of Lagrangian duality =-=[16]-=-. Temporarily relaxing the constraint (6) and applying the Lagrange multiplier 2, then the Lagrangian problem is L = TC + 2 (I- ~ (16) Taking the partial derivative of L with respect to each xi and se... |