P. P. Bhattacharya, L. Georgiadis, P. Tsoucas and I. Viniotis, "Adaptive Lexicographic Optimization in Multi-class M/GI/1 Queues," Mathematics of Operations Research, Vol. 18, No. 3, pp. 705-740, 1993.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....to the work presented in [1] by identifying the workload partitioning problem as a task allocation problem. The discussion of goal orientation presents the relation to some ideas on goal orientation in [11] For an analytical treatment of goal oriented algorithms for the single server case see [2], or [16] The organization of this section was partially inspired by the classification scheme in [25] where classification is based on the a three models approach with a load model, an action model and a solution model plus a separated model for the solution making procedure. 1.1 Primary ....

....sets a (relative) response time goal for each class, and there is an absolute performance vector AP = ap i ) 1in of the same dimension whose entries describe the absolute response time of instances from that class as it was observed in the past. Here average may be given different meanings. In [2] all instances are attributed the same importance, but a weighting according to the order of arrival or to the time of arrival may equally attributed, taking care that the algorithm pursues its goals for a long time successfully. As a result of these two vectors one obtains a third one, a ....

[Article contains additional citation context not shown here]

Bhattacharya P., Georgiadis L., Tsoucas P., Viniotis I.: Adaptive Lexicographic Optimization in Multi-class M/G/1 queues, Mathematics of Operations Research (1992)

....One solution could be to define alternative sets of goals, one more relaxed than the other, and to try to steer the system towards the minimal satisfiable set of goals. Some results are available on this direction in the simple case of a single CPU server with multiple classes of work, see [BGTV92]. Another approach is to endow units of work or classes of them, or even sets of classes (for example, representing work emanating from a specific department or organization) with a budget. Satisfaction of the goals is then relative to the users and their units of work budgets. Utility functions ....

....and system load estimates. FNGD93, FNY93] present dynamic transaction routing algorithms for satisfying administration defined response time goals in shared nothing transaction processing systems. Satisfaction of performance goals for workload classes has also been considered for CPU scheduling [BGTV90, BGTV92], and also for memory management [BCL93, BMCL94, CFW 95] The routing algorithms of [FNGD93, FNY93] take into consideration factors like CPU load of the processing systems, data affinity, transaction priorities, and goal satisfaction of average response time goals for transaction classes, taking ....

P. Bhattacharya, L. Georgiadis, P. Tsoucas, and I. Viniotis. "Adaptive Lexicographic Optimization in Multi-class M/GI/1 queues". Mathematics of Operations Research, 1992.

....desired performance. Or, for each cost term, one defines a basic one theme algorithm, which achieves optimal performance, and derives the state variables from the themes used for achieving the minimization. For a discussion of a transposed version of the first approach, the reader is referred to [1]. The results given in [1] apply mutatis mutandis to some extent to our situation, because user classes may be interpreted as cost function terms. The second approach relies on the identification of optimal algorithms for individual cost terms, and has been investigated in [5] Before we present ....

....for each cost term, one defines a basic one theme algorithm, which achieves optimal performance, and derives the state variables from the themes used for achieving the minimization. For a discussion of a transposed version of the first approach, the reader is referred to [1] The results given in [1] apply mutatis mutandis to some extent to our situation, because user classes may be interpreted as cost function terms. The second approach relies on the identification of optimal algorithms for individual cost terms, and has been investigated in [5] Before we present the results obtained there, ....

Bhattacharya P., Georgiadis L., Tsoucas P., Viniotis I.: Adaptive Lexicographic Optimization in Multi-class M/G/1 queues, Mathematics of Operations Research (1992)

....For each buffer pool and each goal, we define a Performance Index as the measured response time R, divided by 1 Random references are sets of buffer pool requests that are uniformly distributed over all possible buffers and are independent of each other. the goal response time G, see also [9, 2]: P I = R G : The performance goal is being met if the performance index is less than or equal to 1. We define a performance index vector as PI = PI 1 ; P I 2 ; P I M : where PI i is the performance index of buffer pool i; i = 1; M . Defining optimality requires having a ....

....This definition has the drawback that it is only concerned about the worst goal satisfaction. If two different performance index vectors have the same maximal performance index, then we should compare the next highest performance index. This leads to the definition of Lexicographical Comparison, [2]. Assume that we have two performance index vectors P I 0 and PI 1 . To perform a lexicographic comparison, each vector is internally sorted so that PI 1 PI 2 : PIM . Then, the smallest k such that P I 0 k 6= PI 1 k is found. If P I 0 k PI 1 k , then PI 0 is ....

P. Bhattacharya, L. Georgiadis, P. Tsoucas, and I. Viniotis. Adaptive lexicographic optimization in multi-class M/GI/1 queues. Mathematics of Operations Research, 1992.

....DEBIT CREDIT transactions might be in one class while all teller submitted transactions might be in another. Partitioning the transaction workload into disjoint transaction classes and specifying class performance goals is extremely common in commercial on line transaction processing OLTP [1, 3, 6, 5]. The systems where transactions are processed are heterogeneous in nature. They may have different processing speeds, memory and disk resources, and operating systems. The systems also have differing, dynamically varying processing loads. In such an environment, transactions of the same class ....

....arriving transaction is first assigned to a transaction class based on user ID, transaction ID, submitting terminal, etc. The router monitors the transaction arrival rate per class; the arrival rate of class C i is L(i) Each transaction class has an administration defined response time goal G(i) [1, 3, 5, 6]. The database system administrator defines the rules for assigning arriving transactions to transaction classes, and the performance goals for each transaction class. The arriving transaction is passed to the router, which makes a routing decision and updates system state information. The ....

P.P. Bhattacharya, L. Georgiadis, P. Tsoucas, and I. Viniotis. Adaptive lexicographic optimization in multi-class M/GI/1 queues. Mathematics of Operations Research, 18(3):705--740, August 1993.

No context found.

P. P. Bhattacharya, L. Georgiadis, P. Tsoucas and I. Viniotis, "Adaptive Lexicographic Optimization in Multi-class M/GI/1 Queues," Mathematics of Operations Research, Vol. 18, No. 3, pp. 705-740, 1993.

No context found.

P.P. Bhattacharya, L. Georgiadis, P. Tsoucas, and I. Viniotis. Adaptive lexicographic optimization in multi-class M/GI/1 queues with feedback. Mathematics of Operations Research, 18(3):705740, August 1993.

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