L. Rudolph, M. Slivkin-Allalouf, and E. Upfal. A simple load balancing scheme for task allocation in parallel machines. In Proceedings of the 3rd Annual ACM Symposium on Parallel Algorithms and Architectures (SPAA '91), pages 237--245, July 1991.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....projects such as SETI home [23] use software clients that are, in essence, single purpose placeholders that pull work on demand from a server. Placeholder scheduling shares many similarities with self scheduling tasks within a parallel application and the well known master worker paradigm [22], in which placeholders are analogous to worker processes. Of course, our presentation of placeholder scheduling is in the context of job scheduling and not task scheduling. Nonetheless, the basic strategies are identical. Of course, there is a large body of research in the area of job scheduling ....

L. Rudolph, M. Slivkin-Allalouf, and E. Upfal. A Simple Load Balancing Scheme for Task Allocation In Parallel Machines. In Proceedings of the 3rd Annual ACM Symposium on Parallel Algorithms and Architectures, pages 237-245, Hilton Head, South Carolina, U.S.A, July 21-24 1991. ACM Press.

....projects such as SETI home [18] use software clients that are, in essence, single purpose placeholders that pull work on demand from a server. Placeholder scheduling shares many similarities with self scheduling tasks within a parallel application and the well known master worker paradigm [17], in which placeholders are analogous to worker processes. Of course, our presentation of placeholder scheduling is in the context of job scheduling and not task scheduling. Nonetheless, the basic strategies are identical. Of course, there is a large body of research in the area of job scheduling ....

L. Rudolph, M. Slivkin-Allalouf, and E. Upfal. A Simple Load Balancing Scheme for Task Allocation In Parallel Machines. In Proceedings of the 3rd Annual ACM Symposium on Parallel Algorithms and Architectures, pages 237--245, Hilton Head, South Carolina, U.S.A, July 21-- 24 1991. ACM Press.

....of dynamic computations. In contrast to our research on multithreaded computations, however, other theoretical research has tended to ignore space requirements and communication costs. Related work in this area includes a randomized work stealing algorithm for load balancing independent jobs [89]; algorithms for dynamically embedding trees in fixedconnection networks [5, 71] and algorithms for backtrack search and branch and bound [61, 65, 75, 86, 109] Backtrack search can be viewed as a multithreaded computation with no synchronization, and in the work just cited, the only algorithm ....

Larry Rudolph, Miriam Slivkin-Allalouf, and Eli Upfal. A simple load balancing scheme for task allocation in parallel machines. In Proceedings of the Third Annual ACM Symposium on Parallel Algorithms and Architectures, pages 237--245, Hilton Head, South Carolina, July 1991.

....calling convention introduces latent parallelism in the code that can be uncovered by a load balancing mechanism. The load balancing scheme implemented for this benchmark uses two mechanisms: workstealing and thread pushing. Work stealing has been seen before in systems such as Cilk [BL94] and [RSAU91] thread pushing is the dynamic work scheduling problem. An overview of dynamic work scheduling algorithms and techniques can be found in [SHK95] and in [XL97] The metric used to determine a processor s load for load balancing purposes is the time, Tw , that the currently running thread spent ....

Larry Rudolph, Miriam Slivkin-Allalouf, and Eli Upfal. A simple load balancing scheme for task allocation in parallel machines. In ACM Symposium on Parallel Algorithms and Architectures, pages 237--245, 1991.

.... in multi processor computing require processors to balance their actions in some way.Typical examples of such balancing problems include assigning successive memory addresses to processors [18] balancing the computational load on a computer system while minimizing the maximum load on a server [6,35,37,39], and implementing barrier data structures in order to synchronize processes operating at di erent speeds [1, 24, 30,32] In a seminal paper, Aspnes et al. 5] proposed balancing networks as a new approach to solving balancing problems. Balancing networks, resembling comparator networks (see, ....

L. Rudolph, M. Slivkin and E. Upfal, \A Simple Load Balancing Scheme for Task Allocation in Parallel Machines," Proceedings of the 3rdAnnual ACM Symposium on Parallel Algorithms and Architectures, pp. 237-245, July 1991.

....in the system. Furthermore, he shows that for any time interval of fixed length, the maximum system load is likely to be at most log log n= log d O(1) In [VDK96] Vvedenskaya, Dobrushin and Karpelevich independently present similar results. For related results, see [Mit96b, MRS01, Mit97] In [RSAU91] Rudolph, Slivkin Allalouf and Upfal present a simple distributed load balancing strategy. They consider a work generation model in which, at every time step, the load change of any processor due to local generation and service is bounded by some constant. They show that the expected load of any ....

Larry Rudolph, Miriam Slivkin-Allalouf, and Eli Upfal. A simple load balancing scheme for task allocation in parallel machines. In Proceedings of the 3rd Symposium on Parallel Algorithms and Architectures (SPAA '91), pages 237--245, 1991.

.... as a directed acyclic graph [BL94, ABP98] or the tasks may be generated by a backtrack search or branch and bound algorithm [KZ93] The situation where the tasks are independent is somewhat simpler and several models for generating and consuming independent tasks are considered in the literature [RSAU91, BFM98, BFS99] In the random load model each processor in each step generates a task with a xed probability and consumes a task with a xed probability , where 0 1. Whereas in the adversarial load model , the load of each processor at each time can be modi ed arbitrarily by an ....

L. Rudolph, M. Slivkin-Allalouf, and E. Upfal. A simple load balancing scheme for task allocation in parallel machines. In Proceedings of the Third Annual ACM Symposium on Parallel Algorithms and Architectures, pages 237-245, July 1991.

.... as a directed acyclic graph [BL94, ABP98] or the tasks may be generated by a backtrack search or branch and bound algorithm [KZ93] The situation where the tasks are independent is somewhat simpler and several models for generating and consuming independent tasks are considered in the literature [RSAU91, BFM98, BFS99] In the random load model each processor in each step generates a task with a fixed probability and consumes a task with a fixed probability , where 0 1. Whereas in the adversarial load model , the load of each processor at each time can be modified arbitrarily by an ....

L. Rudolph, M. Slivkin-Allalouf, and E. Upfal. A simple load balancing scheme for task allocation in parallel machines. In Proceedings of the Third Annual ACM Symposium on Parallel Algorithms and Architectures, pages 237--245, July 1991.

....lightweight threads packages written for shared memory machines. In particular, we are interested in implementing a scheduler that efficiently supports dynamic and irregular parallelism. 2. 1 Scheduling lightweight threads A variety of lightweight, user level threads systems have been developed [6, 11, 14, 15, 25, 29, 33, 37, 40, 45, 53], including mechanisms to provide coordination between the kernel and the user level threads library [2, 49, 31] Although the main goal of the threads schedulers in previous systems has been to achieve good load balancing and or locality, a large body of work has also focused on developing ....

L. Rudolph, M. Slivkin-Allalouf, and E. Upfal. A simple load balancing scheme for task allocation in parallel machines. In ACM-SIGACT; ACM-SIGARCH, editor, Proceedings of the 3rd Annual ACM Symposium on Parallel Algorithms and Architectures, pages 237--245, Hilton Head, SC, July 1991. ACM Press.

....in multi processor computing require processors to balance their actions in some way. Typical examples of such balancing problems include assigning successive memory addresses to processors [18] balancing the computational load on a computer system while minimizing the maximum load on a server [6, 35, 37, 39], and implementing barrier data structures in order to synchronize processes operating at di erent speeds [1, 24, 30, 32] In a seminal paper, Aspnes et al. 5] proposed balancing networks as a new approach to solving balancing problems. Balancing networks, resembling comparator networks (see, ....

L. Rudolph, M. Slivkin and E. Upfal, \A Simple Load Balancing Scheme for Task Allocation in Parallel Machines," Proceedings of the 3rd Annual ACM Symposium on Parallel Algorithms and Architectures, pp. 237-245, July 1991.

....stealing. Moreover, since each dependency accounts for at most a constant number of messages, the communication cost of resolving any dependency is amortized by the cost n d S max . Thus, Proposition 4.4 immediately implies the stated claims. 2 42 5 Related Work Substantial research (see e.g. [1, 34, 50, 56]) has been reported in the literature concerning the scheduling of multithreaded computations, ignoring though space requirements and communication costs. The work stealing paradigm dates back at least as far as Burton and Sleep s research [16] on parallel execution of functional programs and ....

L. Rudolph, M. Slivkin-Allalouf and E. Upfal, \A simple load balancing scheme for task allocation in parallel machines," Proceedings of the 3rd Annual ACM Symposium on Parallel Algorithms and Architectures, pp. 237-245, Hilton Head, California, July 1991.

....in multi processor computing require processors to balance their actions in some way. Typical examples of such balancing problems include assigning successive memory addresses to processors [18] balancing the computational load on a computer system while minimizing the maximum load on a server [6, 35, 37, 39], and implementing barrier data structures in order to synchronize processes operating at different speeds [1, 24, 30, 32] In a seminal paper, Aspnes et al. 5] proposed balancing networks as a new approach to solving balancing problems. Balancing networks, resembling comparator networks (see, ....

L. Rudolph, M. Slivkin and E. Upfal, "A Simple Load Balancing Scheme for Task Allocation in Parallel Machines," Proceedings of the 3rd Annual ACM Symposium on Parallel Algorithms and Architectures, pp. 237--245, July 1991.

....prove that for any ffl 0, with probablity at least 1 Gamma ffl the algorithm s execution time is O(T 1 =P hT1 log P log(1=ffl) while its communication complexity is O(P (hT1 log(1=ffl) 1 n d )S max ) with probability again at least 1 Gamma ffl. Substantial research (see e.g. [1, 20, 24, 28]) has been reported in the literature concerning the scheduling of multithreaded computations, ignoring though space requirements and communication costs. Burton shows in [10] how to limit space in certain parallel computations without causing deadlock. More recently, Burton [9] has developed and ....

L. Rudolph, M. Slivkin-Allalouf and E. Upfal, "A simple load balancing scheme for task allocation in parallel machines," Proceedings of the 3rd Annual ACM Symposium on Parallel Algorithms and Architectures, pp. 237--245, Hilton Head, California, July 1991.

....by a thread that competes with other threads. On PEs with low load this thread will be scheduled often, so these PEs will spend much of their time trying to find more work. On highly loaded PEs the load balancing thread will seldom be scheduled, allowing them to concentrate on useful work [492]. The pairing for balancing in this scheme is random. This reduces the cost of communication to maintain the required load information, and achieves good results with high probability. Other schemes that use randomization to reduce the cost of collecting load information have also been proposed ....

....PEs get more new threads than PEs that 14 In principle, these queues contain threads from separate applications. In practice, the MAXI implementation was only capable of running a single application at a time, because memory management was not built into the system. 56 already have a high load [492]. Once mapped by get work, threads do not migrate. Similar schemes are used in the KSR1 (see Section 7.2.5) and other systems [507, 414] A variant is used on the Cedar multiprocessor, where the unit of allocation is not a single thread but rather a task with eight threads, that executes on a ....

L. Rudolph, M. Slivkin-Allalouf, and E. Upfal, "A simple load balancing scheme for task allocation in parallel machines". In 3rd Symp. Parallel Algorithms & Architectures, pp. 237-- 245, Jul 1991.

....by a thread that competes with other threads. On PEs with low load this thread will be scheduled often, so these PEs will spend much of their time trying to find more work. On highly loaded PEs the load balancing thread will seldom be scheduled, allowing them to concentrate on useful work [298]. The pairing for balancing in this scheme is random. This reduces the cost of communication to maintain the required load information, and achieves good results with high probability. Other schemes that use randomization to reduce the cost of collecting load information have also been proposed ....

....system, a global queue is used for work distribution, and then local queues are used for the actual execution 11 . The mapping of new threads is done by a system get work thread that competes with local threads, so lightly loaded PEs get more new threads than PEs that already have a high load [298]. Once mapped by get work, threads do not migrate. Similar schemes are used in the KSR1 and other systems [304, 252] A variant is used on the Cedar multiprocessor, where the unit of allocation is not a single thread but rather a task with eight threads, that executes on a cluster of eight PEs ....

L. Rudolph, M. Slivkin-Allalouf, and E. Upfal, "A simple load balancing scheme for task allocation in parallel machines". In 3rd Symp. Parallel Algorithms & Architectures, pp. 237-- 245, Jul 1991.

....when the number of clients is 100. The LS CSRTDBS completes almost 10 more transactions successfully than the CS RTDBS. In a real time environment, this is a very sizeable improvement. 6 Related Work Many techniques have been suggested for load sharing in shared memory multi processor machines [5, 7, 17, 18]. Since the data is visible to all processors, the load sharing exclusively concerns CPU loads. Real time systems have also exploited available information about the submitted tasks so as to guarantee their completion. This information includes the task s deadline, its expected processing time and ....

L. Rudolph, M. Slivkin-Allalouf, and E. Upfal. A Simple Load Balancing Scheme for Task Allocation in Parallel Machines. In ACM Symposium on Parallel Algorithms and Architectures, 1991.

....process: in the case of a stationary distribution, the maximum load of any bin is less than log log(n) log(d) O(1) w.h.p. where n is both the number of balls and bins in the system. In [CS97] the rate of convergence of the infinite process to the stationary distribution is improved. In [RSU91] Rudolph, Slivkin Allalouf, and Upfal present a simple strategy to equalize the load of two processors in one step and show that the expected load of any processor at any point of time is within a constant factor of the average load. They assume a load generation model where at each time step the ....

L. Rudolph, M. Slivkin-Allalouf, and E. Upfal. A simple load balancing scheme for task allocation in parallel machines. In Proceedings of the 3rd Annual ACM Symposium on Parallel Algorithms and Architectures, pages 237--245. ACM Press, 1991.

No context found.

L. Rudolph, M. Slivkin-Allalouf, and E. Upfal. A simple load balancing scheme for task allocation in parallel machines. In Proceedings of the 3rd Annual ACM Symposium on Parallel Algorithms and Architectures (SPAA '91), pages 237--245, July 1991.

....is still small relative to the total execution time. Note that the frequency with which the get work task executes is inversely proportional to the number of envelopes on the local task queue. Thus, it will be scheduled more often on lightly loaded processors, causing them to take more work [4]. Performance can be improved even more, in certain cases. Instead of creating new envelopes each time get work receives control, it first checks the running time of the last activity executing within the envelope. If it started recently, i.e. not more than a time quantum, it does not create a ....

L. Rudolph, M. Slivkin-Allalouf, and E. Upfal, "A simple load balancing scheme for task allocation in parallel machines". In 3rd Symp. Parallel Algorithms & Architectures, pp. 237--245, Jul 1991.

....rescheduling. This effect is countered to a certain degree by affinity scheduling, which attempts to re schedule threads on the same processors they used before [47] Alternatively, it is possible to provide global load balance with local queues and occasional migration to overcome any imbalance [39]. A third problem with a global queue is that the threads in a single application are scheduled in an uncoordinated manner. This may be fine if the threads do not interact with each other. But in many parallel programs the threads do interact and synchronize with each other. If the interaction is ....

L. Rudolph, M. Slivkin-Allalouf, and E. Upfal, "A simple load balancing scheme for task allocation in parallel machines". In 3rd Symp. Parallel Algorithms & Architectures, pp. 237--245, Jul 1991.

No context found.

L. Rudolph, M. Slivkin-Allalouf, and E. Upfal. A simple load balancing scheme for task allocation in parallel machines. In Proceedings of the 3rd Annual ACM Symposium on Parallel Algorithms and Architectures (SPAA), 1991, pages 237-- 245.

No context found.

L. Rudolph, M. Slivkin-Allalouf, and E. Upfal. A simple load balancing scheme for task allocation in parallel machines. In ACM Symposium on Parallel Algorithms and Architectures, pages 237--245, 1991.

No context found.

L. Rudolph, M. Slivkin-Allalouf, and E. Upfal. A simple load balancing scheme for task allocation in parallel machines. In Proceedings of the Second Annual ACM Symposium on Parallel Algorithms and Architectures, pages 237--245, 1991.

No context found.

L. Rudolph, M. Slivkin-Allalouf, and E. Upfal. A Simple Load Balancing Scheme for Task Allocation in Parallel Machines. In ACM Symposium on Parallel Algorithms and Architectures, 1991.

No context found.

L. Rudolph, M. Slivkin-Allalouf, and E. Upfal. A simple load balancing scheme for task allocation in parallel machines. In Proceedings of the Annual ACM Symposium on Parallel Algorithms and Architectures, 1991.

First 50 documents  Next 50

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