Date: Thu, 19 Sep 1996 17:48:53 -0400 (EDT) From: "John T. Farmer" <jfarmer@sabre.goldsword.com> To: jgreco@brasil.moneng.mei.com, rls@mail.id.net Cc: dennis@etinc.com, isp@freebsd.org, jfarmer@goldsword.com, nik@blueberry.co.uk Subject: Re: Routers - hardware received wisdom Message-ID: <199609192148.RAA23438@sabre.goldsword.com>
next in thread | raw e-mail | index | archive | help
Spoken by jgreco@brasil.moneng.mei.com on Thu Sep 19 11:30:01 1996: >Quoting rls@mail.id.net (Robert Shady): >> Let's think about this logically people.. We're only talking about a MAX >> of 187 Kilo-Bytes per second for a single T1 line... I've got calculators >> that could max that out! Now.. When you start throwing multiple ethernet >> devices in there, and you want to provide wire-to-wire speed acrossed those, >> that is another story.. We're using a 486DX4-120Mhz w/32MB of RAM here, and >> it is running 3 100Mb Intel Etherexpress cards, and 2 10Mb SMC Elite Ultra >> cards.. It does a decent job, although I don't know that I would expect to >> be able to get full wire speeds on all ethernet cards simultaneously.. But >> luckily, we have enough segments and switches that we don't need to worry >> about that, yet. > >Rob, > >With all due respect it is not that simple. > >I suspect that with MTU-sized packets, I can easily go wire to wire >with 10baseT at peak speeds even on a 386DX/40 with SMC ISA cards. >Actually I was doing that at one point, IIRC, and it worked fine. > >I suspect that with very small packets, the same machine will have >abysmal performance. > >Dennis' T1 sync serial cards are most similar to an Ethernet card, and >I will flat out state that I can saturate your DX4/120 CPU before I hit >T1 saturation if I attempt to saturate that T1 link with miniature packets. >I have saturated a DX5/133 with this test and it is ugly. > >On the other hand, the router on the other end was clearly swamped and >was only returning one packet for every three I sent (I could see it >on the CSU/DSU lights, it was not due to my local CPU being saturated). >Some Livingston piece of junk, I believe. > >It is clearly very dependent on the kind of data you send. I think I can >floor even a large Cisco with the right kind of abuse so maybe it's a >pointless discussion. I will certainly be the first in line to say that >I was pretty happy with a 386DX/40 as a T1-Ethernet router... but I will >also be the first to properly qualify that statement. True! True! Joe is exactly right! Of course, if I had remembered a tech report that I wrote 10 years ago, I wouldn't had to asked my question... Ancient history (those who are easily bored can leave the room...). About 1987, I was contracted to the local DOE site (Oak Ridge National Labs) to do some studies on ethernet bridging via a broadband CATV network. We were testing some company's (since disappeared) and boxes from Bridge Communications (now a part of 3com, the basis of the NetBuilder series). The boxes were simple multibus, 68000@8MHz, Lance chipset ethernet, feeding a 5mb/sec channel on the CATV net. We used Vax's & Decnet to create protocol loads and a couple of ethernet analyzers (lanalyzer & a _nice_ HP one) to monitor & to do raw loading. By varying the packet sizes, I could saturate the 5mb link, one of the other bridge, or the local ethernet, again depending on the size of packets. (I, of course used this data & some "fancy" statistical functions to show what would happen in the real world with a varied packet stream...) If a 68000/8Mhz box could do that, then the 386/33 is not so unreasonable after all. Granted, I wouldn't go _buy_ one for the task, but since it's idle... John ------------------------------------------------------------------------- John T. Farmer Proprietor, GoldSword Systems jfarmer@goldsword.com Public Internet Access in East Tennessee dial-in (423)470-9953 for info, e-mail to info@goldsword.com Network Design, Internet Services & Servers, Consulting
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199609192148.RAA23438>