[chrissnell]

Great memories. During this era, I was at Citysearch and spent a lot of time in the Orange County Exodus DC. We had a nice rack of Sun gear, Extreme Networks switches and Alteon AceDirector load balancers.

I remember vividly that there was a cage down the row from us that was populated entirely by eMachines, which were a low end desktop PC that you could buy at Circuit City and Best Buy. We laughed at their cage but the company, 911gifts.com, ended up getting acquired for a nice sum, while our site and company was basically gone a few years later.

[alexhawdon]

That's a great story. It'd be interesting to know whether that was all they could afford, or if they were being very forward-thinking in using COTS hardware and a failure-tolerant architecture.

[beachstartup]

probably a combination of both.

i worked at a startup in the late 90's and the price of sun microsystems gear was mind-blowingly high. from memory: ultra 5 workstation with a scsi card and disk array was $15,000+. an ultra e450 server was on the order of $100,000 and went up from there depending on how you wanted to build.

of course, that was exactly the time people started switching to linux on x86 en masse. pentium pro's were good and cheap enough to scale out less expensively on a per unit basis. today you can buy a 72-core xeon server with gigabytes of memory and terabytes of ssd's and 4x 10G ethernet for less than 10 grand. amortized over its functional lifetime, it costs less than a cell phone bill.

[kyledrake]

I got the impression "hardware dogma" still happens when I was planning our infrastructure. The datacenter world is still a place where you can spend $36k for something you can get for $1000 if you're not careful. I didn't use Cisco for our switches and some IT guys I talked to acted like I was a heretic. But the decision saved us several thousand dollars and we ended up with a design that was IMHO more reliable than stacking proprietary switches.

The big trick is to make everything redundant (redundant power, network bonding, Ceph instead of a NAS) and not have a SPOF. Then it matters a lot less if anything fails, and you can use cheaper hardware if you need to. That said, I still prefer server grade hardware - just not always the newest or the most brand name.

[technofiend]

Unfortunately storage dogma still seems to be a thing for EMC shops. Just try suggesting Ceph over Scale I/O. :-)

[technofiend]

I was the technical architect for an internet-based commodity trading system that was also hosted at Exodus in Austin before the .com bust. As I recall we spent $6 million on our two Sun clusters "Coke" and "Pepsi" but that was almost 15 years ago so they might have been $6 million each. Then the bottom fell out of energy trading thanks to Enron and others and that was that. Those systems were probably sold for pennies on the dollar and would easily be replaced today with a small AWS presence for even less.

Frankly that's part of what makes working now such fun: the fact these systems just keep getting better and better.

[scurvy]

Their revenue mattered a lot more than their COGS. At least in this case.