4 ms·
Do you have any performance metrics for this assertion? My understanding was that Netflix and others have contributed very performant code.
by shrubble 2y ago
Do you have any performance metrics for this assertion? My understanding was that Netflix and others have contributed very performant code.
- toast0 2y agoI haven't seen an apples to apples network performance test in a while. Maybe some shops pick their OS based on network performance, but most pick their OS for other reasons and then bend it until the network performance meets their needs. Most OSes have at least good enough network performance; I wouldn't run a public tcp server on MacOS because they don't have syncookies, and I wouldn't run a large multicore server on OpenBSD because afaik, they're missing cpu pinning and I don't know if their scheduler is biased towards keeping processes on the same core --- everything else should be fine. Netflix has contributed some general network performance increases, but the real big increases are only there if you are serving files from disk, with TLS, and you have a NIC that can accelerate the bulk crypto. I like FreeBSD and choose it when I can, but I've never had occasion to benchmark it. I prefer the stability of user experience and the expectation that old documentation still applies that FreeBSD has and Linux doesn't. When I've dug into kernel source, I feel like FreeBSD source is better organized and easier to understand, but that might be familiarity bias.
- epcoa 2y agoUltimately it becomes just as silly as any microbenchmarks. Most people don’t care that they can do 100G line rate. But the superior QoS implementations available for Linux make a real performance difference for many people. For example, as a typical gateway, Openwrt will perform better than pfsense simply because of cake. I am not Netflix, I do not have their problems and very few do. Nevermind their competitors doing fine with Linux.