Per-node multithreading and remote latency

Thitikamol, K.; Keleher, Peter J.

doi:10.1109/12.675711

Cited by 16 publications

(8 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In order to achieve high CPU resource utilization, multithreading [6,17,22] is generally supported in most modern software DSM systems. It allows their users to create multiple threads on each processor to overlap the CPU computation time and network communication time.…”

Section: Performance Model Analysismentioning

confidence: 99%

“…However, network speed and the amount of network traffic have great effect on the performance of programs in a DSM system. Consequently, most of the researches in the past focused on the design of performance-enhancing technologies such as relaxed consistency protocols [2,7,8,16,23], multithreading [6,17,22], load balance [24][25][26], communication minimization [19], etc.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Finding a suitable system scale to optimize program performance on software DSM systems

et al. 2006

View full text Add to dashboard Cite

Recently, software distributed shared memory systems have successfully provided an easy user interface to parallel user applications on distributed systems. In order to prompt program performance, most of DSM systems usually were greedy to utilize all of available processors in a computer network to execute user programs. However, using more processors to execute programs cannot necessarily guarantee to obtain better program performance. The overhead of paralleling programs is increased by the addition in the number of processors used for program execution. If the performance gain from program parallel cannot compensate for the overhead, increasing the number of execution processors will result in performance degradation and resource waste. In this paper, we proposed a mechanism to dynamically find a suitable system scale to optimize performance for DSM applications according to run-time information. The experimental results show that the proposed mechanism can precisely predict the processor number that will result in the best performance and then effectively optimize the performance of the test applications by adapting system scale according to the predicted result.

show abstract

Section: Performance Model Analysismentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Finding a suitable system scale to optimize program performance on software DSM systems

et al. 2006

View full text Add to dashboard Cite

show abstract

“…Since the amount of computation for the application program is fixed, it is important to minimize the delay due to page faults, lock waiting, and barrier waiting to obtain proper performance from the DSM system. An approach alleviating this problem has been proposed in [18] and [19]. Their research is based on using multi-threading so that another thread can execute while a thread is blocked.…”

Section: Introductionmentioning

confidence: 98%

An asynchronous protocol for release consistent distributed shared memory systems

Yeo

Yeom

Park

2000

Proceedings of the 2000 ACM Symposium on Applied Computing - Volume 2

View full text Add to dashboard Cite

Distributed shared memory (DSM) systems[l] provide a simple programming paradigm for networks of workstations, which are gaining popularity due to their cost-effective high computing power. However, DSM systems usually exhibit poor performance due to the large communication delay between the nodes. A lot of different memory consistency models have been proposed to mask the delay due to network communication.In this paper, we propose an asynchronous protocol for the release consistent memory model, which we call Asynchronous Release Consistency (ARC) protocol. Unlike other protocols where the communication adheres to synchronous request/receive paradigm, our protocol is asynchronous in nature such that the necessary pages are broadcast before they are requested. Hence, the network delay is reduced by proper prefetching of necessary pages. We have compared the performance of the proposed protocol with the lazy invalidate protocol by running standard benchmark programs and we found that ARC achieves up to 29% performance improvement.

show abstract

“…Figure 2 shows application performance as the number of threads per processor is increased from 1 to 8 threads per node on conf-hom. The slight increases in performance at small numbers of threads are due to latency hiding [9,10]. Note that this increase would be much larger if we took startup costs into account, as in [10], but we consider only steady-state execution here.…”

Section: Load Balancingmentioning

confidence: 99%

“…Note that this increase would be much larger if we took startup costs into account, as in [10], but we consider only steady-state execution here. Additionally, we are not currently restructuring applications to avoid the problems discussed by Thitikamol [9], such as per-thread reductions, and duplication of thread state. These numbers impose a limit on the number of threads that we can use before performance degrades.…”

Section: Load Balancingmentioning

confidence: 99%