C. Severance, R. Enbody, and P. Petersen, "Managing the overall balance of operating system threads on a multiprocessor using automatic self-allocating threads (ASAT)". J. Parallel & Distributed Comput. 37(1), pp. 106--112, Aug 1996. 16  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... the allocation at runtime to reflect changes in the load and requirements [10] Severance et al. propose a scheme that is less dependent on explicit information, in which the system measures the performance of a barrier synchronization to decide if the current number of threads is appropriate [16]. The closest scheme to ours was proposed by Nguyen et al. who measure the efficiency of a parallel job on several partition sizes and then decide on the allocation [12] However, these schemes involve learning about a specific application at run time, and are irrelevant for other jobs. They do ....

C. Severance, R. Enbody, and P. Petersen, "Managing the overall balance of operating system threads on a multiprocessor using automatic self-allocating threads (ASAT)". J. Parallel & Distributed Comput. 37(1), pp. 106--112, Aug 1996. 16

....and checks the total number of virtual PEs in the system. Jobs that are found to have too many virtual PEs are notified of this condition, and correct it at their convenience [581] Alternatively the jobs may correct the number of threads by themselves, based on performance measurements [521]. Eventually the total number of virtual PEs becomes equal to the number of physical PEs, thus preventing any overhead due to context switching and synchronization delays. This scheme reduces the coordination that is required in cases where the requirements change at a high rate, at the expense of ....

.... that the frequency of reconfigurations be kept low in practice: otherwise the price of reconfiguration might even outweigh the benefits of changing the PE allocation [446, 171, 543] Alternatively, a model where reconfigurations are only allowed at certain points in the application can be used [630, 629, 185, 521]. These points are chosen such that repartitioning is significantly easier than at other points in the computation, e.g. at the beginning of a new parallel loop 8 . Setting the partition sizes The main reason to change the partition sizes at runtime is a desire to improve fairness on the one ....

[Article contains additional citation context not shown here]

C. Severance, R. Enbody, and P. Petersen, "Managing the overall balance of operating system threads on a multiprocessor using automatic self-allocating threads (ASAT)". J. Parallel & Distributed Comput. 37(1), pp. 106--112, Aug 1996.

.... that the frequency of reconfigurations be kept low in practice: otherwise the price of reconfiguration might even outweigh the benefits of changing the PE allocation [272, 97, 327] Alternatively, a model where reconfigurations are only allowed at certain points in the application can be used [365, 364, 109, 315]. These points are chosen such that repartitioning is significantly easier than at other points in the computation, e.g. at the beginning of a new parallel loop 6 . 6 Acquisition of PEs at the beginning of parallel loops is common practice on Cray vector multiprocessors. This is typically done ....

....then execute on available PEs. If there are too many threads in the system, the jobs voluntarily reduce the number of threads. If there are free PEs, the jobs create new threads to utilize them. Examples of this approach include process control [342] and ASAT (Automatic Self Allocating Threads) [315]. A more explicit form of single PE allocation has been taken in a couple of coarse grain systems, which are designed for a distributed programming style more than for a parallel one. The scheduling mechanism is simply a global queue of independent threads. The difference between this and the ....

C. Severance, R. Enbody, and P. Petersen, "Managing the overall balance of operating system threads on a multiprocessor using automatic self-allocating threads (ASAT)". J. Parallel & Distributed Comput. 37(1), pp. 106--112, Aug 1996.

.... the allocation at runtime to reflect changes in the load and requirements [10] Severance et al. propose a scheme that is less dependent on explicit information, in which the system measures the performance of a barrier synchronization to decide if the current number of threads is appropriate [16]. The closest scheme to ours was proposed by Nguyen et al. who measure the efficiency of a parallel job on several partition sizes and then decide on the allocation [12] However, these schemes involve learning about a specific application at run time, and are irrelevant for other jobs. They do ....

C. Severance, R. Enbody, and P. Petersen, "Managing the overall balance of operating system threads on a multiprocessor using automatic self-allocating threads (ASAT)". J. Parallel & Distributed Comput. 37(1), pp. 106--112, Aug 1996. 15

....than active threads across the entire system, it adds a thread. In this way, ASAT jobs always move toward overall system equilibrium by implicitly reacting to any combination of innumerable changes in system load. A number of different approaches to determine the relative thread balance were tried [SevEn94, SevEn94a ,SevEn95, SevEn96]. The approach which was the most successful was to periodically run a barrier synchronization and time the barrier passage. It will be shown later that that there is up to three orders of magnitude difference in barrier passage times when comparing thread balanced to thread imbalanced conditions. ....

.... ported to the Convex Exemplar [CONEXMP] and those results were reported in [SevEn94a] The results of ASAT on the SGI and Convex Exemplar were presented as a poster session in [SevEn94b] The ASAT performance on the Convex Exemplar was compared to the Convex ASAP hardware solution in [SevEn95] In [SevEn96] the second version of ASAT was evaluated on the SGI Challenge. ASAT is currently integrated into a Beta version of a production compiler [Guide] run time environment from Kuck and Associates. The performance of the Guide compiler implementation of ASAT when multiple processes are executed is ....

[Article contains additional citation context not shown here]

Severance C, Enbody R, Petersen P, "Managing the Overall Balance of Operating System Threads on a Multiprocessor using Automatic Self-Allocating Threads (ASAT), to appear in Journal of Parallel and Distributed Computing (JPDC) Special Issue on Multithreading for Multiprocessors.

....detected parallelism include: Automatic Self Adjusting Processors (ASAP) from Convex [1] and Autotasking on Cray Research [2] computers. A previous study of the benefits of Automatic SelfAllocating Threads (ASAT) for the Convex Exemplar was done in [6] details on multiple ASAT jobs appears in [7]. ASAT The general goal of our Automatic Self Allocating Threads (ASAT) is to eliminate thread imbalance by detecting thrashing and then dynamically reducing the number of active threads to achieve balanced execution over the long term. In this way, multi threaded applications will experience ....

Severance C, Enbody R, Peterson P, "Managing the Overall Balance of Operating System Threads on a MultiProcessor using Automatic Self-Allocating Threads (ASAT)," Journal of Parallel and Distributed Computing Special Issue on Multithreading on Multiprocessors, to appear.

....iteration scheduling, and operating system support. Scheduler Activations [1] and Process Control [7] are somewhat similar to Autotasking in that they rely on an agreement between the operating system and the run time library in the multi threaded task. Automatic Self Adjusting Threads (ASAT) [5,6] and Loop Level Process Control (LLPC) 8] do not depend on the operating system for notification about the load condition of the system. Both approaches actively track the load of the system and adjust their threads as appropriate. The primary difference between ASAT and LLPC is the way in which ....

Severance C, Enbody R, Petersen P, "Managing the Overall Balance of Operating System Threads on a Multiprocessor using Automatic Self-Allocating Threads (ASAT), to appear in Journal of Parallel and Distributed Computing (JPDC) Special Issue on Multithreading for Multiprocessors.

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