Date: Mon, 18 Nov 1996 10:13:17 -0600 (CST) From: Joe Greco <jgreco@brasil.moneng.mei.com> To: dennis@etinc.com (dennis) Cc: jgreco@brasil.moneng.mei.com, isp@freebsd.org Subject: Re: changed to: Frac T3? Message-ID: <199611181613.KAA02381@brasil.moneng.mei.com> In-Reply-To: <199611181601.LAA28913@etinc.com> from "dennis" at Nov 18, 96 11:01:07 am
next in thread | previous in thread | raw e-mail | index | archive | help
Hi Dennis, > >I did, briefly, already. > > > >System: ASUS P/I-P55T2P4, P100 CPU, 16MB RAM, Znyx 314 quad DE21040 > > ethernet, NE2000 ethernet. > > Well...a verbal description of what you were doing was what I was looking > for..... Trying to melt down a production router. (Not too sure that's too smart, but oh well). > flood pings are a pretty stinky test....the goal is to find out what it can > switch on a > continuous basis without packet loss...you may be chugging along at 10,000pps > and someone pings the machine and you drop a flurry of packets while the > ping is being processed, which is clearly unacceptable. its also a bit > different with I do not think a router routes a ping packet any differently than it routes a TCP packet or a UDP packet. In both cases, it comes in one interface, gets routed, and goes out another interface. ping -f just happens to be a convenient traffic generator. My "udpblast" program is a little more optimized to the task simply because it generates packets to the discard port. That is how I got around the collision problem inherent on a 10baseT network. Actually I wish the damn things ran in full duplex mode because the two interfaces in question are both directly connected via crossover cables, I should "never" get a collision on them to the machines that they connect to :-/ > lots of different addresses being looked up (rather than the same one with > a ping).....its pretty difficult to test. Sorry, I just don't have the test environment to do that. There is a low level of widely variant traffic, but it accounted for less than 10% of the traffic (recall that I am doing this on production equipment). Since I only have a T1 out to the big net, that sorta limits me a little. > at 350 > > >> Anyone have a feel for the avg packet size over a typical backbone > >> link? A T3 with an avg packet size of 500 bytes is 21000pps full > >> duplex...I suspect the ave packet size may be smaller with lots > >> of dialup traffic..... > > > >I tend to see an average of about 350 bytes. > > ok, thats about 28,000pps (allowing for overhead)...14,000 in > one direction with a full pipe. So if I can do 10,000pps on a P100, possibly more... hmm.. :-) ... JG
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199611181613.KAA02381>