Jump to Content
[go: nahoru, domu]

Networking

Network Performance Decoded: Benchmarking TCP and UDP bulk flows

June 22, 2024
Sumit Singh

Senior Product Manager

Rick Jones

Network Software Engineer

Try Gemini 1.5 models

Google's most advanced multimodal models in Vertex AI

Try it

Here on the Google Cloud networking team, we've been deep in the trenches with our customers, helping them build, fix, and supercharge their networks for years. In that time, we've uncovered some valuable patterns and best practices for maximizing network performance and efficiency. This wealth of knowledge is more than just a theoretical resource — it's a practical toolkit designed to help you achieve your business goals, whether you're deploying on Google Cloud, cross-cloud, on-premises, or on other cloud providers. We want to share this expertise to help you avoid costly mistakes that can drain resources and waste valuable time. 

With this goal, we are excited to announce a series of whitepapers on best practices for network performance and benchmarking, with the first set of whitepapers available today. The knowledge in these whitepapers is especially important in the context of Cross-Cloud Network, where your end-to-end connectivity spans multiple clouds; among other factors, the different cloud-specific Linux distributions, different VM bandwidth caps, etc., can drastically impact performance. 

What are we publishing today?

For the first installment, we’re highlighting a collection of whitepapers that cover best practices for benchmarking Transmission Control Protocol (TCP) and User Datagram Protocol (UDP) bulk flows. Configuring bulk flows is essential for optimizing your network performance. But what does good look like? Whether you are dealing with TCP or UDP for your network data transport, these whitepapers will help you properly configure bulk flows in your environment. 

Considerations When Benchmarking TCP Bulk Flows

Sometimes, a TCP data transfer benchmark between two endpoints in a network exhibits lower-than-expected throughput. Why? And what can you do about it? This whitepaper highlights some common situations where tweaking the Linux networking stack’s default settings can help improve data transfer performance of TCP bulk flows.

Considerations When Benchmarking UDP Bulk Flows

The companion to the above-mentioned TCP Bulk Flows article, this whitepaper highlights some important considerations when benchmarking UDP bulk flow performance between two systems. If you’ve ever run a UDP data transfer benchmark and seen lower-than-expected throughput, this whitepaper is for you.  

Measuring Cloud Network Performance with PerfKit Benchmarker

In addition to informative whitepapers on performance and benchmarking best practices, we’ve also recently updated our whitepaper on Measuring cloud network performance with PerfKit Benchmarker. PerfKit Benchmarker (PKB), you’ll recall, is our open-source toolkit for cloud performance benchmarking that helps you conduct reproducible and comprehensive network performance benchmarking yourself, faster and more easily than you could with conventional performance benchmarking tools. PKB handles network setup, VM provisioning, test execution, and result collection and packaging, making your testing more efficient. Plus, it integrates with BigQuery for powerful automated reporting and analysis.

PKB can help you make apples-to-apples performance comparisons between service providers and take data-backed decisions for choosing cloud providers, deploying your workloads and scaling your existing deployments. PKB is being leveraged in both industry and academia for performance benchmarking. We continue to support the growth of the PKB community and encourage community contributions to the PerfKit Benchmarker codebase. Check out our updated edition of this classic whitepaper here. 

More network performance insights coming your way

These resources are a part of our commitment to network performance transparency, enabling open-source tools, and an open knowledge ecosystem for network performance benchmarking. And while the examples used in these whitepapers are from Google Cloud, the ideas are not Google Cloud-specific, and will help you even for your deployments outside Google Cloud.

You can find all our current and future whitepapers on this webpage.

We hope you find these resources helpful in your network performance optimization and troubleshooting needs. Stay tuned for news of more network performance whitepapers in an upcoming Network Performance Decoded blog post!

Posted in