>* Antonios Anastasiadis <[EMAIL PROTECTED]> [2005-05-21 11:27]: >> Are all the xl-based cards crap without exceptions?
>yes. While I don't doubt Henning knows much much more than I do about such things, this answer doesn't exactly satisfy me. Poor performance on xl nics has been discussed many times on misc, but I have never seen much beyond 'they suck, use xyz'. From my experience, xl performance on FreeBSD, NetBSD, Loonix, and Windows* isn't all that terrible and I've been at a loss to explain why it is so disproportionately bad on OpenBSD. If someone told me that 3com wanted an NDA before giving the inside scoop on their chipsets and that is why the OpenBSD driver is so poor I could understand, but I've never seen anything like that. Running a diff of FreeBSD's xl driver to OpenBSD shows that they diverged long ago and have quite a few differences. ie: FreeBSD uses polling so I wasn't even going to attempt to analyze the other differences between driver sets. (I didn't compare NetBSD's xl driver which would have been a better comparison, but you get the idea) After all that, I'd love to hear more on exactly WHY xl nics suck. Another similar performance issue that I saw here a while back was poor performance of OpenBSD->OpenBSD network transfers (the thread was about NFS write performance I think). When you mix up transfers between other OS's (ie free->open, open->free, linux->open, free->linux, etc ) you don't see that performance loss. I don't remember how that thread ended, but I think it was mostly tips on how to increase NFS write performance and no mention of the network drivers/stack or why OpenBSD->OpenBSD sometimes is a little slow. I have quite a few older Dell GXwhatevers in my lab that come with integrated xl nics and I hate to have to replace the nics in them. but that's just my ranting... Matt
