Comments on: 3 serious hardware questions

Question: 3 serious hardware questions

sylloge — Sun, 21 Dec 2003 19:51:07 -0800

Three'fer:
(1) Is there a significant difference between running Linux on Opterons vs. Xeons? Is one clearly better? For MySQL? For Apache/PHP?
(2) Ever hosted with Cogentco? Is there anyone else with prices like that? (100Mbps at $1000/month)
(3) What's your favorite server vendor for typical 2-way 1U systems? (Looking at Dell, IBM, Penguin; where does Google buy them again?)

By: websavvy

websavvy — Sun, 21 Dec 2003 21:08:10 -0800

Cogent has a very poor reputation. You can search for them at Webhostingtalk, which is where those who host, and those who buy hosting, gather to complain and to praise. Cogent's bandwidth is 3rd tier in quality.

By: majick

majick — Sun, 21 Dec 2003 23:01:15 -0800

Hiya Stewart! Getting ready for the big build-out?

1. Opterons support 64-bit memory addressing if you run under a 64-bit kernel and plan to use Big Gigs O RAM. There are some bigger caches available for Opterons, so single-application boxes might be slightly faster for it. If you're going to run a 64 bit distribution, you'll probably find the Opterons to be generally faster. Memory speeds with most of the Xeon boxes are pretty darn good, though. Guessing at what your app is shaped like [ ;) ] I'd say the Opteron advantage for node servers isn't that great, but on the back end it might be worthwhile.

2. Can't speak to the question at all, I'm afraid. We just pull in fiber straight from BBN to each of our data centers.

3. The Dells are OK low-end machines, nothing that'll rock your socks but servicable and affordable. As a vendor, they've swung back and forth between unbeatable support and abyssimal, and I have received DOA hardware from them in the past. The IBM hardware is very well made and easy to work with physically. I haven't used Penguins, but their reputation is good.

By: majick

majick — Mon, 22 Dec 2003 10:56:19 -0800

Also, I don't recall who Google's favorite vendor is, but remember that they built the fairly high failure rate of cocmmodity server hardware into the architecture of the system, so reliability isn't exactly tops on their list of criteria. I recall reading that at any given moment there are hundreds of dead machines in their data centers, and some of them may never actually be fixed because the cost of hunting them down exceeds the value of their function.

Your investors probably won't let you get away with the same strategy (although you could always ask....)