Extrae.jl: Julia bindings for the Extrae HPC Profiler
Sergio Sanchez-Ramirez, Mosè Giordano · 2025
Read Extrae.jl: Julia bindings for the Extrae HPC Profiler by Sergio Sanchez-Ramirez, Mosè Giordano (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-12087.txt.
Introduction
Sergio Sanchez-Ramirez and Mose Giordano link a profiler to the Julia language in a 2025 paper. Julia is popular in high performance computing because code reads like Python but can run near C speed. Squeezing out that speed needs a good profiler, a tool that shows where time goes. Their package, Extrae.jl, connects Julia programs to the Extrae profiler, a long used tool in supercomputing centers. The paper shows how the link works and what Julia users gain from it when tuning large codes.
Catalog record
| CREATOR | Sergio Sanchez-Ramirez, Mosè Giordano |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 19 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-12087
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Extrae.jl: Julia bindings for the Extrae HPC Profiler Sergio Sanchez-Ramirez, Mosè Giordano Abstract The Julia programming language has gained acceptance within the High-Performance Computing (HPC) community due to its ability to tackle two-language problem: Julia code feels as high-level as Python but allows developers to tune it to C-level performance. But to squeeze every drop of performance, Julia needs to integrate with advanced performance analysis tools, also known as profilers. In this work, we present Extrae.jl , a Julia package to interface with the Extrae profiler. keywords Julia, HPC, Profiling, Tracing, Performance Evaluation 1 Introduction Scientific computing applications require code that is both high-level, for expressing complex mathematical ideas, and high-performance, as it usually involves computationally intensive tasks. These requirements derive into a problem known as "the two-language problem"; i.e. programming languages usually focus only on one of these two topics, forcing package authors to use two programming languages: one higher-level for the abstraction and one lower-level for performance. This is characteristic of programming languages such as Python, whose package authors derive to C, C++, or Rust for improving the performance. In recent years, the Julia programming language [ 5 ] has steadily gained adoption within the High-Performance Computing (HPC) community due to its ability to tackle the aforementioned two-language problem: Julia code feels as high-level as Python but allows developers to tune it to C-level performance. Additionally, it supports parallel programming models such as MPI, master-worker distributed computing, and GPU computing, and offers a rich interactive REPL experience, facilitating ease of use and fast prototyping for developers. While being able to write code so that it can exploit computing resources is key, it is not the full story. In pursuit of higher computational power, hardware vendors have made more parts of their systems available to software. This in turn makes software development more relevant than ever for fully exploiting the performance capabilities of current computers. Driving performance optimization are performance analysis tools that are commonly named as profilers . Profilers measure where a program spends its time or which functions consume the most resources. They help identify bottlenecks, optimize code, and improve overall application performance. One of the most well-known profilers in the field is Extrae. Recognized for its low overhead and large support of programming models and platforms, it is a veteran in the performance analysis field. Its programming model allows for deep customization and tuning, but unfortunately, it also imposes a high barrier for newcomers. In this work, we present Extrae.jl : The Julia bindings to the Extrae profiler, with which we try to ease the barrier for Julia users. This work is organized as follows. In Section 2 , we introduce the current state of performance analysis tools in the HPC world to the reader. In Section 3 , we describe Extrae’s design and how we have mapped it Julia’s concepts and parallelism models. In Section 4 , we present some examples of Extrae.jl running on top of Julia HPC apps. Finally, in Section 5 , we summarize our work and point to possible directions of improvement in future works. 2 Background Depending on the target resource of the analysis and the method used to gather the data, different tools are available. Basically, two methods are used for profiling: • Tracing: It involves explicitly calling the profiler at the target regions of the code. Although 100% is accurate, the overhead is high. Manual instrumentation of the code or binary interception of symbols can be used. • Statistical sampling: It entails periodically interrupting the application, collecting data, and resuming. The results are expected to be statistically similar to the ones provided by tracing but with a much lower overhead. There are a number of commercial and open source profiling tools that can be used to analyze performance in HPC based on these two methods. The following is a list of profilers that are known to work with Julia: Julia standard library Profiler.jl Julia comes with a statistical profiler called Profiler.jl , which can be used to periodically sample the backtrace of each currently running task, capturing the name and file-line information of all Julia functions in the call stack. Profiler.jl can also optionally capture information about Julia’s core and, optionally, external shared libraries, written in languages such as C, Fortran, etc. The default output format produced by Profiler.jl is text-based, but external tools can be used to produce various types of visualization. Some of these tools are FlameGraphs.jl , VS Code, ProfileView.jl , ProfileVega.jl , StatProfilerHTML.jl , ProfileSVG.jl , PProf.jl , and ProfileCanvas.jl . It is also possible to profile memory allocations within Julia code with this standard library. Although Profiler.jl does not have native support for distributed computing, it is possible to collect separate traces for different workers. Linux Perf/OProfile Julia has support for analyzing programs using popular open source system profilers for the Linux operating system OProfile, a statistical sampler, and its more modern replacement Perf. By setting the environment variable ENABLE_JITPROFILING=1 Julia will create and register an event listener to profile just-in-time (JIT) compiled functions. Furthermore, the package LinuxPerf.jl [ 12 ] allows getting more fine-grained information about hardware counters using Perf for specified function calls. Nvidia Nsight System The Nvidia Nsight System is a proprietary statistical sampling profiler with tracing features, which supports MPI and CUDA applications. As such, it can be used to profile also Julia programs, including those distributed using the MPI protocol and/or employing a Nvidia GPU, but JIT functions will have their names mangled by the compiler. The package NVTX.jl [ 10 ] allows you to instrument the code and annotate specific functions or code regions, optionally also adding a payload (e.g. the value of a local variable) which can be displayed in the Nsight System GUI. The in-development version of Julia can optionally be compiled to automatically instrument with NVTX internal components, such as the compiler and the garbage collector. Intel VTune Intel VTune is a proprietary sampling profiler with support for MPI and OpenMP applications. When the environment variable ENABLE_JITPROFILING=1 is set, Julia JIT functions will be recorded with their correct name on VTune traces. You can instrument Julia code with the package IntelITT.jl [ 11 ] to get richer information in the traces. LIKWID LIKWID is an open-source suite of tools for performance analysis of HPC applications on Linux, using hardware performance counters, rather than sampling or tracing profiling. It supports MPI, OpenMP, and GPU offloading. The package LIKWID.jl [ 4 ] interfaces with the shared library part of the LIKWID suite to instrument the code and collect important hardware metrics to measure the performance of an application. Score-P Score-P is an open-source suite for profiling and event tracing of HPC applications. You can use the package ScoreP.jl [ 3 ] to instrument Julia code and analyze it with ScoreP. MPITape.jl MPITape.jl [ 1 ] is an experimental open-source Julia package which overdubs MPI operations performed through the package MPI.jl [ 8 ] . This allows you to track communication patterns and timings within a distributed application. Tracy Tracy is an open-source frame and sampling profiler, originally designed for gaming applications but can be used for any other kind of program. Julia can optionally be built