Date: Mon, 30 Dec 1996 16:00:59 -0600 (CST) From: Richard Stanford <richards@herald.net> To: Jacob Suter <jsuter@intrastar.net> Cc: FreeBSD ISP List <isp@freebsd.org> Subject: Re: Bandwidth.. Message-ID: <Pine.A32.3.91.961230155300.143360B-100000@future.dsc.dalsys.com> In-Reply-To: <199612302010.OAA22944@intrastar.net>
next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, 30 Dec 1996, Jacob Suter wrote: > Oh geez no.. I put my current web server (AMD 5x86/133 w/ 32 megs ram) to > the test.. 16,000 hits in 12 hours and it wasn't even really stressed. Is this really a volume test? Sounds kinda low to me. We have a single user's home page that gets around 7,000 hits a day. > > I dont think that anyone would recommend a '386 for anything nowadays > > (with '486-100s with on-board PCI IDE at about $100.)....I was talking > > much more about the '486 or 100Mhz Pentium vs the higher end > > stuff than obsoleted equipment like '386. > Any kind of quality 486 system (even an SX/25) would work well for a web > server on a 128K link pretty much no matter what kind of content > (10-zillion little icons to 40 meg graphics), even on a pretty cheazy > system ($9 hard drive controller). Sure, a 128k link. But considering that P5-133s are, oh, around $200 these days, why not go for them? If you can afford (thinking US here) more than a 64K line, you should be able to get a comfortable webserver (P5, SCSI, et cetera). This is assuming you're reselling webspace (or giving it away w/ dialup accounts, something more than a personal webserver). Is it "needed"? Probably not, on a low-volume link. But it's a cheap comfort, IMO. Just think of it this way -- how many customers would you need to lose through irritation if there's a problem to pay for the hardware? Not /that/ many. The only trouble with things like IDE drives and $9 controllers on a production machine that I see is reliability. Speed is not the issue (not on a 128K link) but moving from a 2% chance per year to a 1% chance per year of failure is more than worth it (fictional numbers, but you get the idea). The point is to make your servers bulletproof and efficient, then you don't have to worry about them. Also, you should get in extra hardware for when they do (and they will) fail .. you can't prevent this, but you can minimize the number of times it happens. -Richard
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.A32.3.91.961230155300.143360B-100000>