These benchmarks on @fastly's Compute@Edge vs Cloudflare workers are seriously flawed.The biggest problem is that they conflate network round trip time with “execution at edge” time. CF is measuring cherry-picked RTT instead of edge compute performance. From their blog:
@fastly Our early numbers show @fastly’s Compute@Edge Rust implementation is much faster than Cloudflare’s published numbers for their JavaScript implementation. We will publish comparisons once we have enough data to be statistically significant, because that is kind of important.
@fastly You have to dig into the details of any benchmark because they are rarely simple, and can easily be constructed to suit specific systems. Source: I spent more than a decade working at a web browser company.
To build a meaningful benchmark a great deal of care has to go into setting up the test correctly.
You have to make sure the test doesn't hit artificial limits (for example, free trial resource limits) as primary evidence suggests happened in this case.
The test methodology Cloudflare used is also flawed, even as an RTT comparison. They didn’t publish which CatchPoint nodes were used so, conveniently, no one else can do apples to apples comparisons.
The benchmarks claim to compare @fastly’s JS implementation which is in beta vs CF’s JS, which is not. I wonder why they didn’t measure Rust which is much more mature on our platform than theirs.
I encourage everyone to run their own benchmarks on @fastly, bearing in mind there are CPU limits on free trial accounts. You won’t be able to benchmark on Cloudflare (and we can’t either) because their terms of service forbid it. fastly.com/signup/
• • •
Missing some Tweet in this thread? You can try to
force a refresh