I. Stoica and H. Abdel-Wahab, "A new approach to implement proportional share resource allocation," Technical Report TR-95-05, CS Dept., ODU, April 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....of sharing a single resource in a fair manner; however, as far as we can tell, none of the singleresource fair queueing schemes suggested in the literature (such as Weighted Fair Queueing [4, 8] Worst case Fair WFQ [3] Start time Fair Queueing [6] etc. nor the proportional share schemes [11, 10, 9] generalize to this multi resource problem detailing the exact reasons why this is so is beyond the scope of this report. 1.3. Proportionate Progress and Pfairness The notion of proportionate progress, and the associated concept of pfairness [1] deals with the following scheduling problem: ....

I. Stoica and H. Abdel-Wahab. A new approach to implement proportional share resource allocation. Technical Report TR--95--05, Department of Computer Science, Old Dominion University, 1995.

....knowledge of task execution times and prohibitions on task interactions. Strict limits are also placed on processor utilization, so that even transient overloads are disallowed. A number of deterministic proportional share scheduling algorithms have recently been proposed [BGP95, FS95, Mah95, SAW95] Several of these techniques [FS95, Mah95, SAW95] make explicit comparisons to lottery scheduling [WW94] although none of them demonstrate support for the higher level abstractions introduced with lottery scheduling. In general, the 1 Anyone who has had the misfortune of trying to implement ....

....on task interactions. Strict limits are also placed on processor utilization, so that even transient overloads are disallowed. A number of deterministic proportional share scheduling algorithms have recently been proposed [BGP95, FS95, Mah95, SAW95] Several of these techniques [FS95, Mah95, SAW95] make explicit comparisons to lottery scheduling [WW94] although none of them demonstrate support for the higher level abstractions introduced with lottery scheduling. In general, the 1 Anyone who has had the misfortune of trying to implement precise scheduling behavior by setting Unix nice ....

[Article contains additional citation context not shown here]

Ion Stoica and Hussein Abdel-Wahab. A new approach to implement proportionalshare resource allocation. Technical Report 95-05, Old Dominion University, Department of Computer Science, Norfolk, Virginia, April 1995.

.... manner; however, as far as we can tell, none of the single resource fair queueing schemes suggested in the literature (such as Weighted Fair Queueing [5] Generalized Processor Sharing [6, 7] Fair Queueing Based on Start times [8] or the schemes in [1, 4, 3] nor the proportional share schemes [9, 10, 11] generalize to this multi resource problem. Detailing the exact reasons why this is so is beyond the scope of this extended abstract: we will, however, attempt to illustrate the problem by means of a simple example: Example 1 Fair queueing and proportional share scheduling algorithms associate ....

I. Stoica and H. Abdel-Wahab. A new approach to implement proportional share resource allocation. Technical Report TR--95--05, Department of Computer Science, Old Dominion University, 1995.

....is said to achieve a lag bound of Delta. Note that a P fair schedule achieves a lag bound of less than 1. The problem of designing an efficient on line algorithm for single resource scheduling that achieves a small lag bound has recently received a great deal of attention in the literature [4, 6, 8, 9, 10]. However, none of the algorithms presented in the aforementioned papers achieves a constant lag bound (i.e. independent of n, the number of tasks) let al..one P fairness. In this paper, we extend a result of Tijdeman [7] to obtain a simple P fair algorithm for scheduling a dynamic set of tasks on ....

I. Stoica and H. Abdel-Wahab. A new approach to implement proportional share resource allocation. Technical Report TR--95--05, Department of Computer Science, Old Dominion University, April 1995.

....recursive application of the basic technique that exhibits improved throughput accuracy and reduced response time variability compared to prior schemes. 7. 2 Proportional Share Schedulers Several other deterministic approaches have recently been proposed for proportional share processor scheduling [Fon95, Mah95, Sto95]. However, all require expensive operations to transform client state in response to dynamic changes. This makes them less attractive than stride scheduling for supporting dynamic or distributed environments. Moreover, although each algorithm is explicitly compared to lottery scheduling, none ....

....support for the flexible resource management abstractions introduced with lottery scheduling. Stoica and Abdel Wahab have devised an interesting scheduler using a deterministic generator that employs a bit reversed counter in place of the random number generator used by lottery scheduling [Sto95]. Their algorithm results in an absolute error for throughput that is O(lg n a ) where n a is the number of allocations. Allocations can be performed efficiently in O(lg n c ) time using a tree based data structure, where n c is the number of clients. However, dynamic modifications to the set of ....

I. Stoica and H. Abdel-Wahab. "A New Approach to Implement Proportional Share Resource Allocation, " Technical Report 95-05, Department of Computer Science, Old Dominion University, Norfolk, VA, April 1995.

No context found.

I. Stoica and H. Abdel-Wahab, "A new approach to implement proportional share resource allocation," Technical Report TR-95-05, CS Dept., ODU, April 1995.

....is said to achieve a lag bound of Delta. Note that a P fair schedule achieves a lag bound of less than 1. The problem of designing an efficient on line algorithm for single resource scheduling that achieves a small lag bound has recently received a great deal of attention in the literature [8,13,16 18]. In the context of scheduling network traffic in packet based networks, Demers, Keshav, and Shenker [7] have proposed the Packet Fair Queueing algorithm (PFQ) Parekh and Gallager [10,11] have analyzed the behavior of PFQ when the traffic conforms to the (oe; ae) model [6] and shown that any ....

I. Stoica and H. Abdel-Wahab. A new approach to implement proportional share resource allocation. Technical Report TR--95--05, Department of Computer Science, Old Dominion University, April 1995.

....operating system should seamlessly integrate these new types of applications with conventional interactive and batch applications. Many schedulers have tried, and in part have succeeded, to address these requirements. Generally, these schedulers fall in two categories: proportional share [2, 26, 28, 29, 30], and real time 1 based schedulers [20, 21, 22] In proportional share algorithms each client has associated a weight which determines the share of the resource that the client should receive. The scheduler tries to allocate the resource among competing clients in proportion to their share. For ....

....range of applications, this might increase the complexity and possibly the scheduling overhead. Although in EEVDF we are not providing similar support for expressing clients priorities, we note that this support can be integrated in higher level resource abstractions such as monetary funds [26]. 11 11 In [26] we consider two classes of services: bounded and unbounded. A client receiving a bounded service is guaranteed to receive a share of the resources which is inferior and or superior bounded. 7.3 Fair Queuing The EEVDF algorithm shares many common characteristics with the fair ....

[Article contains additional citation context not shown here]

I. Stoica, H. Abdel-Wahab, "A new approach to implement proportional share resource allocation", Technical Report TR-95-05, CS Dpt., ODU, April 1995.

....and eliminates the need to keep an event queue. It is worth mentioning that unlike other previous approximations [5] ours guarantees optimal lag bounds. Besides the class of fair queuing algorithms, a significant number of other proportional share algorithms have recently been developed [1, 9, 12, 16]. Although none of them guarantees constant lag bounds in a dynamic system, we note that the PD algorithm of Baruah, Gehrke, and Plaxton [1] achieves constant lag bounds in a static system. The idea of applying fair queueing algorithms to processor scheduling was first suggested by Parekh in ....

I. Stoica, H. Abdel-Wahab, "A new approach to implement proportional share resource allocation", Technical Report TR-95-05, CS Dpt., Old Dominion Univ., April 1995.

....the need of keeping an event queue. It is worth to mention that unlike other previous approximations [5] ours maintains the same tight bounds for the client lag. Besides the class of fair queuing algorithms, recently, a significant number of other proportional share algorithms have been developed [1, 9, 12, 15]. Although none of them guarantees constant lag bounds in a dynamic system, we note that the PD algorithm of Baruah, Gehrke, and Plaxton [1] achieves constant lag bounds in a static system (i.e. a system in which the number of clients does not change) The idea to apply fair queueing algorithms ....

I. Stoica, H. Abdel-Wahab, "A new approach to implement proportional share resource allocation", Technical Report TR-95-05, CS Dpt., ODU, April 1995, pp. 17 (also available at http://www.cs.odu.edu/~stoica/pubs.html)

....a predictable and responsive way. In addition, a general purpose operating system should seamlessly integrate these new types of applications with conventional interactive and batch applications. In the past years, the proportional share schedulers have emerged as a viable solution to this problem [1, 2, 4, 9, 7, 11, 14, 15]. Although, in general, proportional share schedulers offer weaker guarantees for applications with timeliness constraints than the traditional real time based schedulers [6, 7] they tend to be more flexible and ensure a graceful degradation in overload situations. In a proportional share system, ....

....be interrupted, i.e. once started they must complete in the same time quanta. For example, once a communication switch begins to send a packet for one session, it cannot serve any other session until the entire packet is sent. Therefore, similarly to other proportional share allocation algorithms [1, 2, 4, 9, 7, 11, 14, 15], we assume that the resource is allocated in discrete time quanta of size at most q. At the beginning of each time quantum, a client is selected to use the resource. Once the client acquires the resource, it may use it either for the entire time quantum, or it may release it before the time ....

I. Stoica and H. Abdel-Wahab, "A new approach to implement proportional share resource allocation", Technical Report TR-9505, CS Dpt., ODU, April 1995.

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