On Tue, Sep 15, 2009 at 10:02:16PM +0200, frantisek holop wrote: > hmm, on Tue, Sep 15, 2009 at 01:41:01PM -0500, Marco Peereboom said that > > > the devs are sitting around smug saying nothing until someone comes > > > up with this theme again and then they send something like: > > > > > > i have a bgp machine forwarding 800MBit/s of real world generic > > > internet traffic. can handle at least twice that. > > > > > > what can i say? i am happy for you :] > > > > And the person who does that has some sort of vested interest to lie to > > you? And he also has the worst intentions for the community? > > i did not mean that it could be a lie, or fabrication. i meant that > this is not a replacement for benchmarks. why not describe some > high profile situations like this for others to see what is openbsd > capable of? no wonder everybody was asking for details. this is > exactly the (perceived) problem with openbsd's performance: noone > really knows what can be expected of the system. there is precious > little data about this, and most of it outdated as well. Send me a fully licensed spirent and you get your network benchmarks for OpenBSD but I don't have the time nor the money to do a correct setup for benchmarks. I for myself will not publish flawed benchmarks that can't be correctly reproduced. I run tests for myself to check assumptions but not for publication.
> the word benchmark has a lot negative connotations, mostly because > it's 'hard to do it well'. but let's call then benchmarks 'load > testing' and that is not hard to do at all -- push the system until > it falls. everyone would be interested in the results. doesn't > even have to compared to other systems, although that puts it into > even more context as well. > Funnily we try to make system not fall even under massive load. The best example is the dynamic RX DMA tuning and livelock prevention done with the help of MCLGETI(). On good systems packet floods exceeding the forwarding performance of the system will no longer kill or lock the system. Even the userland routing daemons will get enough CPU time to keep the sessions alive. The goals of OpenBSD do not include to be the fastest OS in the world. We try to solve performance issues quickly. The base system should be fast enough for reasonable setups. If it gets un-reasonable be prepared to invest some effort yourself and trust me you realize when you enter the un-resonable part of the scale. -- :wq Claudio
