From owner-freebsd-questions@FreeBSD.ORG Tue Jun 1 08:35:27 2004 Return-Path: Delivered-To: freebsd-questions@freebsd.org Received: from mx1.FreeBSD.org (mx1.freebsd.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id 0CCC816A503 for ; Tue, 1 Jun 2004 08:35:27 -0700 (PDT) Received: from calypso.bi.lt (calypso.bi.lt [213.226.153.10]) by mx1.FreeBSD.org (Postfix) with ESMTP id 6F13B43D45 for ; Tue, 1 Jun 2004 08:35:26 -0700 (PDT) (envelope-from hugle@vkt.lt) Received: by calypso.bi.lt (Postfix, from userid 506) id 8AD96598269; Tue, 1 Jun 2004 18:35:25 +0300 (EEST) X-Original-To: freebsd-questions@freebsd.org Received: from vkt-dell (lan1.vkt.lt [213.226.136.201]) by calypso.bi.lt (Postfix) with ESMTP id CEF05598225; Tue, 1 Jun 2004 18:35:24 +0300 (EEST) Date: Tue, 1 Jun 2004 18:35:11 +0300 From: hugle Organization: VKT X-Priority: 3 (Normal) Message-ID: <1396574573.20040601183511@vkt.lt> To: Bill Moran , freebsd-questions@freebsd.org In-Reply-To: <40BCA063.2010007@potentialtech.com> References: <8935715836.20040531193600@vkt.lt> <20040531142628.E90411-100000@cactus.fi.uba.ar> <553059104.20040601002503@vkt.lt> <40BBBDC9.8090807@potentialtech.com> <17388683440.20040601101847@vkt.lt> <40BC8099.4050007@potentialtech.com> <188538784.20040601165436@vkt.lt> <40BCA063.2010007@potentialtech.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Subject: Re[2]: routing for 1000 users and 10Mbit internet. X-BeenThere: freebsd-questions@freebsd.org X-Mailman-Version: 2.1.1 Precedence: list Reply-To: hugle List-Id: User questions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Tue, 01 Jun 2004 15:35:27 -0000 BM> hugle wrote: >> BM> hugle wrote: >> >>>>BM> hugle wrote: >>>> >>>>>>FG> On Mon, 31 May 2004, hugle wrote: >>>>>> >>>>>>>>dammit.. >>>>>>>>why then my users eats so much CPU? >>>>>>>>look: >>>>>>>>CPU states: 0.0% user, 0.0% nice, 0.8% system, >>>>>>>>38.0% interrupt, 61.2% idle >>>>>>>>Mem: 21M Active, 177M Inact, 133M Wired, 1228K Cache, 199M Buf, 1677M Free >>>>>>>> >>>>>>>>I have only 61% idle ? >>>>>>>>usualy i have ~50 idle.. >>>>>>>>now I have P4 2.4GHZ >>>>>>>> >>>>>>>>maybe my setup is bad (kernel I mean)? >>>>>> >>>>>>FG> My guess is either cheap hardware (NIC) or bad tuning. What NICs are you >>>>>>FG> using? >>>>>> >>>>>>I'm using Intel cards (fxp) at the moment. But from reading the posts >>>>>>I've decided to buy GBIT NIC. >>>>>>Now the dilema is what brand name.. INTEL or 3COM ? maybe you guys >>>>>>could advice? >>>> >>>>BM> I've always had good results with fxp cards. >> >>>>how much users do you have? and what model? >> >> BM> Never more than 100 ... it's been a year or two, so I don't remember the model. >> >>>>>>And yes, my PC is dealing with lots of network traffic.. It's a >>>>>>gateway + shaping. It deals with almost 1000 users.. >>>>>> >>>>>>Will this PC (p4 2.4GHZ) deal with 10mbit internet? as a gateway + >>>>>>shaper ? with one GBIT NIC connected to 100Mbit swith? >>>> >>>>BM> A gbit NIC isn't going to run at gbit speed on a 100mbit switch. >> >>>>I know, but as I know it'll have more memory, buffer or smth like >>>>that. which somehow will help to deal with the problem, right? >> >> BM> Probably. It just seems like a lot of $$$ to drop when you haven't >> BM> tried polling yet. Keep in mind, that if you try polling and it >> BM> doesn't work, you can just turn it back off, and you haven't spent >> BM> any $$$ on hardware that didn't help. >> >>>>BM> I would look elsewhere than the NIC. Intel NICs are good units (in my experience, >>>>BM> if someone knows of problems with them, please speak up) >>>>from systam -v: >>>>481 fxp0 irq12 >>>>226 fxp1 irq3 >>>>317 fxp2 irq7 >>>>I think it is quite high? right? >> >> BM> I guess. I would expect numbers like that considering the load it's >> BM> under. >> >> BM> I'll ask _again_ ... is the machine's performance poor? Fact is, if >> BM> you give it enough network traffic to shape, it's going to raise the >> BM> CPU load, no matter what you do. >> >> now the main problem is.. that machine is shaping internet, right? >> I did shaping for my subnet, so users in /24 have 100kbits everybody. >> But they don't get such speed, they get about 70-80kbps . and if I >> try to skipt pipe rules for certain IPs, users get all available >> speed (which is left), it's about 500kbps.. >> so why machine can't pipe it normally ? >> Didn't had these problems in the past... BM> Past when? What changed? actualy nothing... more users joint the LAN. BM> Typically, only ~80% of available bandwidth is usable. I don't know if BM> that applies to your situation, though, as that's usually referring to BM> ethernet, and you claim the problem hasn't always been there. no.. for example if i start downlaoding without using pipes (no shaping) my total bandwith usage increases up to 98-99% BM> Try polling and see if the load reduces and the performance increases. BM> If all that machine is doing is routing, you can configure it to be BM> dedicated to routing. I'll do it in ~10-12 hours.. I'll plug my server off, plug in new 3com gbit NIC. and boot my kernel with new options - device_POOLING. and see what is happening and so on.. PS. i think my PC has too less CPU. cause I think that NAT + shaper both eats much processor. and read somewhere, that: interrupts apper if CPU doesn't manage to deal with them. So, after changing NIC, i'll post my result here hugle