Date: Sat, 16 Oct 1999 13:25:23 -0500 From: "Jimbo Bahooli" <griffin@blackprojects.org> To: "Mike Smith" <mike@smith.net.au> Cc: freebsd-hackers@freebsd.org Subject: Re: Balancing Outgoing traffic over 2 nics, and nic limitations. Message-ID: <199910161325230440.0DE208AE@207.109.8.249> In-Reply-To: <199910161735.KAA06493@dingo.cdrom.com> References: <199910161735.KAA06493@dingo.cdrom.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On 10/16/99 at 10:35 AM Mike Smith wrote: >> Hello my FreeBSD friends. >> >> I have two issues. >> >> The first is how to balance outbound traffic over 2 nics that are on >> the same subnet. Example configuration: >> >> fxp0: 12.2.2.5 netmask 255.255.255.0 >> fxp1: 12.2.2.6 netmask 255.255.255.255 >> >> router at: 12.2.2.1 > > >You can't do this. If all of the outbound traffic is headed for the >same router, put two cards in the router and use two separate nets. > >> Currently I have the obvious static route to 12.2.2.1, which locks onto >> fxp0 so all outbound traffic flows out over that link. Inbound traffic >> balances per ip as I would expect. I hope to find a scalable solution >> as I hope to build a server that will utilize 3 nics. > >This is not a sensible course of action. > Sounds fair. :) >> This configuration is neccessary because by my estimation I have run >> into a limit on the intel pro 100 netcards of 6,000 packets/second. > >These cards do not exhibit such a limit. You may have run into some >issues with FreeBSD's ability to handle very large numbers of small >packets with your particular application mix. > >> This limit equates to about 30 to 32 megabit/second of web traffic in >> our situation. I am wondering if anyone else has noticed this limit? > >Not in my recollection. The fxp driver in recent incarnations limits >the number of interrupts it generates by restricting them to >low-resource conditions rather than generating one per packet. And >I've personally seen an SMP kernel run tolerably while taking > 100,000 >interrupts per second. > >> This limit was hit on 2 very different machines, one with significantly >> less power. Any feedback on either of these issues would be >> appreciated. > >I'd start by eliminating the network adapter and driver; move to an >up-to-date FreeBSD-stable and substitute a 3C905B or C and determine >for yourselves whether this is really an issue with the card. > >General experience would suggest that you should be able to come close >to saturating your network with even relatively small datagrams using >either of these adapters. > >You also don't mention whether you're running on a switched network; at >that sort of traffic level you will definitely want to be using a >switch that supports full duplex operation. Of course its a switched network with full duplex operation. But now that the general answer is that it is not a limitation of the nic card I am going to look elsewhere. I was not to sure if it was actually a limit myself, its just that I observed it on two different machines. They however were not huge powerhouses, one was a p2-450, and one was a dual p2 333. Both running real new versions of 3.3-stable. Thanks for the replies at any rate. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199910161325230440.0DE208AE>