Date: Sun, 27 Jan 2002 23:50:43 +0000 From: Neil McGann <neil@neilmcgann.co.uk> To: TD790@aol.com Cc: freebsd-questions@freebsd.org Subject: Re: UDMA and 100Mbit NIC speed issues Message-ID: <5.1.0.14.0.20020127233735.00a3ac60@pop.ntlworld.com> In-Reply-To: <15d.7fa5980.2985d379@aol.com>
next in thread | previous in thread | raw e-mail | index | archive | help
At 17:04 27/01/02 -0500, you wrote: >In a message dated 1/27/02 4:03:42 PM Eastern Standard Time, >neil@neilmcgann.co.uk writes: > > > Realistically 100Mb/sec is flat out > > for 33MHz PCI and trying to get some bandwidth for the NIC may just > not be > > do-able. > >Its more like 400Mb/s, but highly dependent on what else is on the bus. >100Mb/s is what you'll get with single quadword transactions, but any >burstable bus-master will be able to do substantially better than that. You >can do routing with dual full duplex 100Mb/s ethernets without seeing xmitter >underruns, which requires 400Mb/s across the bus. A 3rd nic or disk >transactions will cause problems beyond that. Not on a single bus it isn't - 33Mhz, 32-bit width = 132Mbytes/sec theoretical max - less address cycles and general overhead. 66MHz and/or 64-bit is obviously a different story - as are chip sets with multiple parallel PCI data paths. If UDMA100 can really hit 100Mbytes/sec for extended periods then it's going to need serious control of burst sizes and latencies to allow much other concurrent throughput on that physical bus. Neil To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?5.1.0.14.0.20020127233735.00a3ac60>