Skip to content

Latest commit

 

History

History
124 lines (102 loc) · 4.83 KB

README.md

File metadata and controls

124 lines (102 loc) · 4.83 KB

NOTE: The interfaces are being actively developed and are not yet stable. The github issue tracker is the primary way to communicate with the developers.

Build Status Read the Docs Code Style: Black

PerfFlowAspect: a tool to analyze cross-cutting performance concerns of composite scientific workflows.

Overview

High performance computing (HPC) researchers are increasingly introducing and composing disparate workflow-management technologies and components to create scalable end-to-end science workflows. These technologies have generally been developed in isolation and often feature widely varying levels of performance, scalability and interoperability. All things considered, optimizing the end-to-end workflow amidst those considerations is a highly daunting task and thus it requires effective performance analysis techniques and tools.

Unfortunately, there still is a paucity of techniques and tools that can analyze the end-to-end performance of such a composite workflow. While a myriad of analysis tools exist for traditional HPC programming paradigms (e.g., a single application running at scale), there has been a lack of studies and tools to understand the effectiveness and efficiency of this emerging workflow paradigm.

Enter PerfFlowAspect. It is a simple Aspect-Oriented Programming-based tool that can cast a cross-cutting performance-analysis concern or aspect across a heterogeneous set of components (e.g,, combining Maestro and custom workflow pipeline with Flux along with microservices running on on-premises kubernetes machines) used to create a modern-day composite science workflow.

PerfFlowAspect will provide multiple language support most relevant for HPC workflows including Python. It is designed specifically to allow researchers to weave the performance aspect into critical points of execution across many workflow components without having to lose the modularity and uniformity as to how performance is measured and controlled.

Quick Start

perfflow is the python package that contains the PerfFlowAspect tool for the python language. src/python/perfflowaspect/aspect.py contains a key annotating decorator. Users can use the @perfflowaspect.aspect.critical_path() decorator to annotate their functions that are likely to be on the critical path of the workflow's end-to-end performance. These annotated functions then serve as the Join points that can be weaved with PerfFlowAspect to be acted upon.

The following shows a simple snippet that annotates two functions.

import perfflowaspect.aspect

@perfflowaspect.aspect.critical_path()
def bar(message):
    time.sleep(1)
    print(message)

@perfflowaspect.aspect.critical_path()
def foo():
    time.sleep(2)
    bar("hello")

def main():
    foo()

Once annotated, running this python code will produce a performance trace data file named perfflow.<hostname>.<pid>. It uses Chrome Tracing Format in JSON so that it can be loaded into Google Chrome Tracing to render the critical path events on the global tracing timeline. Please see here where Colt McAnlis actually wrote about using Chrome Tracing for game profiling.

Design and mechanics of PerfFlowAspect

When these annotated functions, or join points, are weaved with PerfFlowAspect, we can invoke specific performance-analysis actions, a piece of tracing code, on those points of execution. They are often referred to as advice in Aspect Oriented Programming.

PerfFlowAspect currently supports only one type of advice: ChromeTracingAdvice. src/python/perfflowaspect/advice_chrome.py implements this advice class. This particular advice simply logs a performance event data in CTF.

PerfFlowAspect's annotation also supports the notion of pointcut: a predicate that matches join points. In fact, @perfflowaspect.aspect.critical_path() can take an optional keyword argument called pointcut whose value can be around, before, after or their async variations (around_async, before_async, and after_async). pointcut=around will invoke the advice before and after the execution of the annotated function whereas pointcut=before or pointcut=after will only advise either corresponding point of function execution.

License

SPDX-License-Identifier: LGPL-3.0

LLNL-CODE-764420