Date: Sat, 20 Mar 1999 00:04:22 -0800 From: David Greenman <dg@root.com> To: Amancio Hasty <hasty@rah.star-gate.com> Cc: Matthew Dillon <dillon@apollo.backplane.com>, Julian Elischer <julian@whistle.com>, Bill Paul <wpaul@skynet.ctr.columbia.edu>, hackers@FreeBSD.ORG Subject: Re: Gigabit ethernet revisited Message-ID: <199903200804.AAA12975@implode.root.com> In-Reply-To: Your message of "Fri, 19 Mar 1999 22:06:35 PST." <199903200606.WAA38098@rah.star-gate.com>
next in thread | previous in thread | raw e-mail | index | archive | help
>I think that David Greeman quoted 2 gigabytes/sec memory bandwith for >the Xeon processor with whatever chipset he was using. That was Intel's claim when using 8-way interleaved, 50ns EDO memory (which is how this machine is configured). I think whoever said that at Intel was wrong, however. It doesn't seem to be as fast as even 1 GB/second. I'm actually getting lower transmit numbers than Bill is for the equivilent tests, on a machine that should be much faster than Bill's, and that's another mystery we've been trying to resolve. -DG David Greenman Co-founder/Principal Architect, The FreeBSD Project > > Cheers, > Amancio > >> :Bill Paul wrote: >> :> >> :> The receiving host is under heavy interrupt load. Andrew Gallatin has >> :> said to me that this is a classic case of livelock, where the system >> :> is so busy processing interrupts that nothing else is getting done. >> :> In this case, the NIC is dutifully DMAing all the packets to the host >> :> and the driver is queing them all to ether_input(), but this is happening >> :... >> : >> :I think Andrew might be right.. >> :it could well be livelock. >> : >> :Matt Thomas implemented a solution for the 100mb dec cards >> :when 100 was fast. I think that the de drivers responded to the >> :interrupt and immediatly did SCHEDNETISR() to schedule the rest of >> >> Hey cool, at least the hardware problem has been solved! >> >> On the livelock thingy -- well, one way to find out is to use ipfw >> to throw away the packets, and then do a 'systat -vm 1' to see where >> the cpu's time is being sent and how much cpu is being used. >> >> I'm guessing between 50% and 70% of the cpu is being eaten with the >> packets going into an ipfw bitbucket. >> >> Each memory read or write represents around 85 MBytes/sec. The DMA counts >> as one. The read() system call counts as two ( because it must read from >> one memory location and write to another ) -- this puts us perilously >> close to the memory bandwidth limit of the cpu when you count all the >> other garbage going on that's breaking up the L1 cache. >> >> :julian >> >> -Matt >> Matthew Dillon >> <dillon@backplane.com> >> >> >> >> To Unsubscribe: send mail to majordomo@FreeBSD.org >> with "unsubscribe freebsd-hackers" in the body of the message > > > > >To Unsubscribe: send mail to majordomo@FreeBSD.org >with "unsubscribe freebsd-hackers" in the body of the message 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?199903200804.AAA12975>