Hacker News Reader

Comment by bummer904 | Hacker News Reader

The main thing keeping me from using OpenBSD is that TCP transfers seem to take at least twice as long (testing both small and large files over plain HTTP) as Linux. Has anyone else seen this?

View on HN

I have heard from a guy on IRC that openbsd's biggest flaws are the network stack and file system options compared to freebsd and gnu/linux.

Twice as long! Have you found an existing bug report? Did you remove every difference in your testing except for the OS?

Yes. I'm just wondering if anyone has actually compared side by side and can either corroberate or contradict my findings.

I'm not sure what you mean. Certainly many people would have noticed if OpenBSD's networking was generally half as fast as it's supposed to be. It must be something related to your specific configuration (not necessarily just software).

Actually I think very few OpenBSD people would notice if under load OpenBSD was significantly slower than Linux because very few OpenBSD people deploy in situation where they can do sustained parallel load tests.

Linux has BBR. OpenBSD doesn't have BBR. The difference between BBR and Cubic or classic TCP is immense.

That's a good point. I read the comment such as that every transfer over TCP would be twice as slow, which would be pretty absurd, but the poster probably meant high load/bandwidth situations. It would not surprise me at all if OpenBSD fared way worse, there.

It should be said that OpenBSD is a fine system, and this specific (BBR) difference would only show up if you had significantly large amounts of data to transfer in flows. I suspect normal use of OpenBSD as most people deploy it (which is not streaming data at high speed) wouldn't care.

I have nothing against OpenBSD. If the developers don't want to implement BBR right now because of more important stuff on their backlog, thats fine with me.

It's not exactly the same thing, but I've read posts where people are saying that pfSense (FreeBSD) can't route at gigabit speeds on a PCEngines APU2, while a similar Linux-based routing distribution can on the same hardware.

What findings? Where is the report and methodology?

OpenBSD doesn't support sendfile(2), which might account for performance issues in some cases, but certainly not twice as long.

Do you have a link to the bug report you undoubtedly filed when observing this?

That's... surprising. Any reason why they wouldn't want to have that on OpenBSD?

Reply on news.ycombinator.com