Return to BSD News archive
Newsgroups: comp.os.386bsd.questions Path: sserve!newshost.anu.edu.au!munnari.oz.au!news.Hawaii.Edu!ames!agate!doc.ic.ac.uk!uknet!pipex!sunic!news.chalmers.se!cs.chalmers.se!augustss From: augustss@cs.chalmers.se (Lennart Augustsson) Subject: Re: Is com performance improved under NetBSD 0.9 ? Message-ID: <1993Oct13.163959.7638@cs.chalmers.se> Keywords: n Sender: news@cs.chalmers.se (News Account) Organization: Chalmers University of Technology References: <CEA7HD.2IH@flatlin.ka.sub.org> <CEJ1D4.3y7@pilhuhn.sub.org> <291lpt$q80@homer.cs.mcgill.ca> Date: Wed, 13 Oct 1993 16:39:59 GMT Lines: 32 In article <291lpt$q80@homer.cs.mcgill.ca> storm@cs.mcgill.ca (Marc WANDSCHNEIDER) writes: >In article <CEJ1D4.3y7@pilhuhn.sub.org>, >Heiko W.Rupp <hwr@pilhuhn.sub.org> wrote: >>bad@flatlin.ka.sub.org (Christoph Badura) writes: >>>>I have been using NetBSD 0.8 for some time now, and am very happy with it, >>>>except for one thing: the performance of serial i/o. >> >>>Serial performance has increased a lot with NetBSD-0.9. It's quite >>>possible to run at 38400 bps on a 386/40 and only get a couple of silo >>>overflows. However the system on which we (Heiko and me) do this >> >>We naturally have fifos installed to acheive this - even when running >>two 38400 bps lines at the same time, we don't get too much silo >>overflows. >>Compiling the kernel with -O2 and DUMMY_NOPS also decreased the number >>of overflows. > > I have IDE drives, and I find that when I am doing ANYTHING other > than leaving the downloads alone, I get big time silo overflows > and CRC errors. I had exactly the same problem. Since the sio driver seems to be far off still I've hacked the com driver. I'll post the patches ASAP. It seems to work well (on my 486/33 laptop with some 16450 like UARTs). I've now transferred 30M (I'm supping current) without a single silo overflow and sometimes with lots of disk activity. (The modem is 9600, but I'm running 38400 to the modem.) -- -- Lennart Augustsson [This signature is intentionally left blank.]