Yandex has developed and open-sourced Perforator, a tool designed to identify and evaluate code inefficiencies to aid in server infrastructure optimisation.
Perforator is a resource that enables developers to pinpoint the most resource-intensive code segments and offers statistics for improving these inefficiencies, thus potentially reducing CPU resource usage by up to 20% each year. Depending on the scale of a company, this reduction could lead to financial savings ranging from millions to billions of dollars annually, thereby allowing businesses to redirect resources towards innovation and expansion. The tool is available for free access on GitHub.
Sergey Skvortsov, a Senior Developer at Yandex who heads the team behind this tool, explained, "Perforator helps businesses get the most out of their servers without sacrificing performance. Using Perforator, businesses can optimize their code, reduce server load, and ultimately lower energy and equipment costs."
Resource optimisation is highlighted as essential for both large data centres and burgeoning startups with limited resources. Perforator allows companies to optimise existing infrastructure rather than expand it, effectively supporting businesses in maintaining performance without additional investment. Having been utilised within various Yandex services for over a year, Perforator is now available globally to companies, developers, and researchers.
The tool aids companies in deploying Perforator on their own servers, thereby reducing their dependency on external cloud providers and supporting data security needs. "Perforator can benefit companies of all sizes, from small businesses with 10-100 servers, which can save millions of dollars per year, to larger enterprises with thousands of servers and more, where savings can reach hundreds of millions or even billions of dollars annually," said Skvortsov. "Regardless of your company size, Perforator can help you reduce infrastructure costs, freeing up resources for further innovation and growth."
Perforator offers detailed insights into server resource consumption and the impact of code on performance. Using eBPF technology, it operates small programs within the Linux kernel safely and efficiently without system slowdown. Supporting programming languages such as C, C++, Go, Rust, Python, and Java, Perforator provides data visualisation through flame graphs, simplifying issue diagnostics.
The tool also features profile-guided optimisation (PGO), which can enhance the performance of C++ programmes by as much as 10%. Additionally, Perforator is designed for seamless operation on single computers, broadening its accessibility beyond large enterprises to include startups and tech enthusiasts. Its capabilities also cover A/B testing, allowing businesses to make better-informed decisions.
In a commitment to community collaboration and technological advancement, Yandex has open-sourced Perforator. "We believe that open-sourcing such fundamental system technologies helps drive tech innovation worldwide," stated Skvortsov. "We aim for our technologies to benefit the world and provide value to both developers and businesses. Additionally, the openness of the technology enables us to make decisions regarding the development of the profiling infrastructure together with the community."
Future updates planned for Perforator include enhanced integration with Python and Java and more refined event analysis metrics. This tool joins other Yandex open-source projects on GitHub, such as YaFSDP, a resource aimed at accelerating large language model training.
