Fine-grained multiprocessor real-time locking with improved blocking

Ward, Bryan C.; Anderson, James H.

doi:10.1145/2516821.2516843

Cited by 26 publications

(17 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Thus, any request for a CE lock may be blocked by ρg − 1 other requests. From the blocking analysis of DGLs [22], the total number of blocking requests for a CE is at most ρg − 1. Analytical bounds for P2P and system memory migrations differ.…”

Section: Resultsmentioning

confidence: 99%

“…As shown in [21], worst-case blocking grows quadratically with respect to the total number of GPU tokens if these locks are acquired separately. We avoid such excessive blocking by instead using dynamic group locks (DGLs) [22]. Using DGLs, a job atomically requests both CE locks simultaneously, instead requesting of one after the other.…”

Section: B Gpusync Structurementioning

confidence: 99%

See 1 more Smart Citation

Exploring the Multitude of Real-Time Multi-GPU Configurations

Elliott

Anderson

2014

2014 IEEE Real-Time Systems Symposium

Self Cite

View full text Add to dashboard Cite

Abstract-Motivated by computational capacity and power efficiency, techniques for integrating graphics processing units (GPUs) into real-time systems have become an active area of research. While much of this work has focused on single-GPU systems, multiple GPUs may be used for further benefits. Similar to CPUs in multiprocessor systems, GPUs in multi-GPU systems may be managed using partitioned, clustered, or global methods, independent of CPU organization. This gives rise to many combinations of CPU/GPU organizational methods that, when combined with additional GPU management options, results in thousands of "reasonable" configuration choices. In this paper, we explore real-time schedulability of several categories of configurations for multiprocessor, multi-GPU systems that are possible under GPUSync, a recently proposed highly configurable real-time GPU management framework. Our analysis includes the careful consideration of GPU-related overheads. We show system configuration strongly affects realtime schedulability. We also identify which configurations offer the best schedulability in order to guide the implementation of GPU-based real-time systems and future research.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: B Gpusync Structurementioning

confidence: 99%

Exploring the Multitude of Real-Time Multi-GPU Configurations

Elliott

Anderson

2014

2014 IEEE Real-Time Systems Symposium

Self Cite

View full text Add to dashboard Cite

show abstract

“…SPEPP [26] and M-BWI [15], [19] apply the notion of a helping mechanism, where a waiting task can execute on behalf of the preempted resource holder. More recently, a protocol named RNLP [28], [27] has been developed to support nested resource requests by applying a token mechanism and a set of request satisfaction mechanisms that can fit into different system models. In [18], a new multiprocessor task partitioning and resource allocating algorithm is proposed to offer a guaranteed speedup of 11-6/(m+1), where m is the number of processors in the system.…”

Section: B Related Workmentioning

confidence: 99%

New schedulability analysis for MrsP

Zhao

Garrido

Burns

et al. 2017

2017 IEEE 23rd International Conference on Embedded and Real-Time Computing Systems and Applications (RTCSA)

View full text Add to dashboard Cite

Abstract-In this paper we consider a spin-based multiprocessor locking protocol, named the Multiprocessor resource sharing Protocol (MrsP). MrsP adopts a helping-mechanism where the preempted resource holder can migrate. The original schedulability analysis of MrsP carries considerable pessimism as it has been developed assuming limited knowledge of the resource usage for each remote task. In this paper new MrsP schedulability analysis is developed that takes into account such knowledge to provide a less pessimistic analysis than that of the original analysis. Our experiments show that, theoretically, the new analysis offers better (at least identical) schedulability than the FIFO non-preemptive protocol, and can outperform FIFO preemptive spin locks under systems with either intensive resource contention or long critical sections.The paper also develops analysis to include the overhead of MrsP's helping mechanism. Although MrsP's helping mechanism theoretically increases schedulability, our evaluation shows that this increase may be negated when the overheads of migrations are taken into account. To mitigate this, we have modified the MrsP protocol to introduce a short non-preemptive section following migration. Our experiments demonstrate that with migration cost, MrsP may not be favourable for short critical sections but provides a better schedulability than other FIFO spin-based protocols when long critical sections are applied.

show abstract

“…As shown in the full version of this paper [11], if these locks are acquired separately, then worst-case blocking grows quadratically with respect to the total number of GPU tokens. We avoid such excessive blocking by instead using dynamic group locks (DGLs) [26]. Using DGLs, a job atomically requests both copy engine locks simultaneously.…”

Section: Migrationsmentioning

confidence: 99%

“…Thus, any request for a copy engine lock may be blocked by ρg − 1 other requests. From the blocking analysis of DGLs [26], the total number of blocking requests for the copy engine is at most (ρg − 1). Since no task requires more than X max time to complete b…”

Section: P2p Imentioning

confidence: 99%

GPUSync: A Framework for Real-Time GPU Management

Elliott

Ward²,

Anderson³

2013

2013 IEEE 34th Real-Time Systems Symposium

Self Cite

110

View full text Add to dashboard Cite

The integration of graphics processing units (GPUs) into real-time systems has recently become an active area of research. However, prior research on this topic has failed to produce real-time GPU allocation methods that fully exploit the available parallelism in GPU-enabled systems. In this paper, a GPU management framework called GPUSync is described that enables increased parallelism while providing predictable real-time behavior. GPUSync can be applied in multi-GPU real-time systems, is cognizant of the system bus architecture, and fully exposes the parallelism offered by modern GPUs, even when closed-source GPU drivers are used. Schedulability tests presented herein that incorporate empirically measured overheads, including those due to bus contention, demonstrate that GPUSync offers real-time predictability and performs well in practice.

show abstract

Fine-grained multiprocessor real-time locking with improved blocking

Cited by 26 publications

References 12 publications

Exploring the Multitude of Real-Time Multi-GPU Configurations

Exploring the Multitude of Real-Time Multi-GPU Configurations

New schedulability analysis for MrsP

GPUSync: A Framework for Real-Time GPU Management

Contact Info

Product

Resources

About