Profiling of Dataflow Programs Using Post Mortem Causation Traces

Brunet, Simone Casale; Mattavelli, Marco; Janneck, Jörn W.

doi:10.1109/sips.2012.54

Cited by 16 publications

(13 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…the longest, time-weighted sequence of events from the start of the program to its termination. Some works [13] [3] investigate the evaluation of the critical path using post-mortem causation trace at high abstraction-level. One of them [3] has even proposed an algorithm for evaluating the critical path in a linear time.…”

Section: Related Workmentioning

confidence: 99%

“…Some works [13] [3] investigate the evaluation of the critical path using post-mortem causation trace at high abstraction-level. One of them [3] has even proposed an algorithm for evaluating the critical path in a linear time. But, contrary to our approach, the need of the causation trace which can be very complex prevents its use at runtime scheduling.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Towards run-time actor mapping of dynamic dataflow programs onto multi-core platforms

Yviquel

Casseau

Raulet

et al. 2013

2013 8th International Symposium on Image and Signal Processing and Analysis (ISPA)

View full text Add to dashboard Cite

The emergence of massively parallel architectures, along with the necessity of new parallel programming models, has revived the interest on dataflow programming due to its ability to express concurrency. Although dynamic dataflow programming can be considered as a flexible approach for the development of scalable applications, there are still some open problems in concern of their execution. In this paper, we propose a low-cost mapping methodology to map dynamic dataflow programs over any multi-core platform. Our approach finds interesting mapping solutions in few milliseconds that makes it doable at regular time by translating it in an equivalent graph partitioning problem. Consequently, a good load balancing over the targeted platform can be maintained even with such unpredictable applications. We conduct experiments across three MPEG video decoders, including one based on the new High Efficiency Video Coding standard. Those dataflow-based video decoders are executed on two different platform: A desktop multi-core processor, and an embedded platform composed of interconnected and tiny Very Long Instruction Word -style processors. Our entire design flow is based on open-source tools. We present the influence of the number of processors on the performance and show that our method obtains a maximum decoding rate for 16 processors.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Towards run-time actor mapping of dynamic dataflow programs onto multi-core platforms

Yviquel

Casseau

Raulet

et al. 2013

2013 8th International Symposium on Image and Signal Processing and Analysis (ISPA)

View full text Add to dashboard Cite

show abstract

“…Our runtime simulator is not a complete CAL execution engine, but it is based on a CAL simulator and profiler which, for a given input, provides accurate information referred to as a causation trace [2], [7]. The causation trace is a directed acyclic graph (DAG) describing firings of actions, dependencies between them along with a plethora of profiling information, such as the number of consumed tokens on each buffer and the actual number/type of operations carried out in each action.…”

Section: A the Setupmentioning

confidence: 99%

“…Note that we view these two actors as the environment, and map them onto processors that always run at the maximum speed, while their impact on the energy consumption is ignored. The causation trace used by our simulator was obtained for three frames of the Foreman benchmark sequence, and contains a DAG with over 75000 nodes (action firings) [2].…”

Section: A the Setupmentioning

confidence: 99%

See 1 more Smart Citation

Runtime voltage/frequency scaling for energy-aware streaming applications

Gruian

2012

2012 Conference Record of the Forty Sixth Asilomar Conference on Signals, Systems and Computers (ASILOMAR)

View full text Add to dashboard Cite

Power and energy consumption, today essential in all types of systems, can be reduced by scaling the voltage/frequency at runtime and/or powering down idle components. Efficient management requires not only pertinent decisions, but also early access to workload information, as well as domain specific solutions. This paper focuses on runtime energy management for streaming applications running on multiprocessor platforms with dynamic voltage/frequency (speed) scaling capabilities. Our energy management occurs at processor level, and employs a number of orthogonal techniques based on hints gathered from load history, buffer pressure and future workload estimates. The manager operates both through speed adjustments and priorities. A preliminary evaluation based on a high-level simulation of an MPEG-4 SP decoder, shows that a combination of specific and generic techniques is the closest to an ideal energy lower bound.

show abstract

Tracing and Profiling Machine Learning Dataflow Applications on GPU

Zins

Dagenais

2019

Int J Parallel Prog

View full text Add to dashboard Cite

In this paper, we propose a profiling and tracing method for dataflow applications with GPU acceleration. Dataflow models can be represented by graphs and are widely used in many domains like signal processing or machine learning. Within the graph, the data flows along the edges, and the nodes correspond to the computing units that process the data. To accelerate the execution, some co-processing units, like GPUs, are often used for computing intensive nodes. The work in this paper aims at providing useful information about the execution of the dataflow graph on the available hardware, in order to understand and possibly improve the performance. The collected traces include low level information about the CPU, from the Linux Kernel, as well as high-level information about the dataflow model. This is followed by post-mortem analysis and visualization steps in order to enhance the trace and show useful information to the user. To demonstrate the effectiveness of the method, it was evaluated for TensorFlow, a well-known machine learning library that uses a dataflow computational graph to represent the algorithms. We present a few examples of machine learning applications that can be optimized with the help of the information provided by our proposed method.

show abstract

Profiling of Dataflow Programs Using Post Mortem Causation Traces

Cited by 16 publications

References 14 publications

Towards run-time actor mapping of dynamic dataflow programs onto multi-core platforms

Towards run-time actor mapping of dynamic dataflow programs onto multi-core platforms

Runtime voltage/frequency scaling for energy-aware streaming applications

Tracing and Profiling Machine Learning Dataflow Applications on GPU

Contact Info

Product

Resources

About