April 24, 2006
One of the key differences between the original dot-com bubble and the Web 2.0 bubble we're entering now is that our servers are a lot cheaper and a lot more powerful. Moore's Law in action isn't exactly news, but the new web is definitely powered by cheap "whatever boxes":
In the 10 years between Excite and JotSpot, hardware has literally become 100X cheaper. It's two factors Ã¢â‚¬â€œ Moore's law and the rise of Linux as an operating system designed to run on generic hardware. Back in the Excite days, we had to buy proprietary Sun hardware and Sun hard drive arrays. Believe me, none of it was cheap. Today, we buy generic Intel boxes provided by one of a million different suppliers.
We recently specced out a new server at work and I was curious about this: exactly how much more powerful did servers get in the last six years?
The parts list for our homebrew server is saved in a newegg wishlist. I set out to find a year 2000 equivalent by looking up a typical Dell server on the mid-2000 internet archive of Dell's website. According to that page, an entry-level PowerEdge 4400 server started at $4,814. I can't get to the detailed spec pages, so I'm estimating the entry-level specs based on the many PowerEdge 4400 machines for sale on eBay.
|Typical 2006 server
||Typical 2000 server
|Dual Core 64-bit CPU, 2.0 Ghz
||Two 32-bit CPUs, 733 Mhz
|4 GB DDR400 memory
||512mb PC133 ECC memory
|150 GB 10,000 RPM SATA-II mirrored
||9 GB 10,000 RPM UltraSCSI mirrored
Now, this comparison isn't entirely fair. The PowerEdge 4400 supports real hot-swappable power supplies and hardware RAID-ed hard drives; our homebrew rig has to be powered down to switch out a failed hard drive or (single) power supply.
But the general thrust of the comparison is still valid. In a nutshell, we get..
- 10x the network bandwidth
- 8x the memory
- 4x the memory bandwidth
- 16x the disk space
- 10x the CPU power
.. all for about one-third the price. And we have the luxury of running a commonly available 64-bit operating system in native 64-bit mode, too.
I would still argue that the the relative costs of software and hardware-- relative to the cost of human labor, I mean-- haven't changed that much in the last six years. But you can plainly see where this extravagant excess of server power makes it possible to use labor-saving software that wasn't viable in the year 2000. You can build your site on extremely high-level software, such as interpreted languages like Ruby and Python, and still scale across thousands of simultaneous user requests.. even on a single "whatever box" server.
Posted by Jeff Atwood
Since I'm such an intently lazy person, could I perchance to wonder (rather than bother to research) how powerful a server could one buy from Dell right now at a similar price with similar redundancies and hotswappabilities? You've more than made your point anyway, but it'd be nice. And secondhand boxen from auctions are an absolute joy for learning hobbyists. :)
You will always have the "Are we ready to take the next step" situation.
I actually think it is very few situations where people take the step to early. People always do a lot of research on new machines and technology. But maybe too many take the step too late?
Ether way it is a technical decision that are funny and exciting; switching to new technology and new ground.
But the problem is not what you solve your problems with, but how you solve them.
My manager can easily drag’n’drop together applications from his UML diagram faster then I can enter my console. It will on his new machine perform and he can show it off and give me a display on how worthless my knowledge in pointers and garbage collection is.
But sooner or later the application will start to leak, the database connected has grown too large or some other issue.
I do not fear adopting higher levels of languages but I fear the careless coding.
If you want to compare to scsi fairly, just add $600-800 for the difference. Even still, it's half the price.
It is shocking how different the tech world is from the time when NT 4 and SunOS 5 were king, especially when it comes to management of networks. I occasionally find myself thinking, it's a shame they only had 1 GHz ten years ago - until I remember that it was only 6! :p
The captcha really must be done in green, or blue, or any other color, perhaps a random one. Or maybe I'm just goofing off too much.
how powerful a server could one buy from Dell right now at a similar price with similar redundancies and hotswappabilities?
There were two reasons we opted to build our own:
1) Dell has an unholy pact with Intel, so we'd be stuck with Intel Inside. And Intel's Xeon CPUs are currently quite inferior to the AMD Opteron/Athlon64.
2) The Dell machine was about 2x the price for similar (but more hot-swappable) stuff. I'd guesstimate $4k+, possibly even $5k.
any other color
I've been approaching this the other way in my datacenter; I've been buying up (off ebay, natch) as many $100 servers as I can. This buys me dual PIII@900mhz, 1GB ram, dual scsi HDs. This is the box that used to cost $4K not so many years ago- and at these costs, I can afford to buy ten and make them redundant- and since everything gets filtered by a 100Mhz feed anyway (slightly slower than a 1x CDrom drive), it pretty much keeps up. So I can run yesterday's eminently suitable hardware- at 2.5% of the cost.
I just checked the Dell site and the closest 2006 config to our $1,750 newegg homebrew is the PowerEdge 2850 in "Enhanced" config:
- Dual Xeon 3.0 GHz
- Dual 74GB 10k SCSI
- 4GB RAM
- dual 1000baseT network
As I said, expensive, and the P4 Xeons underperform the AMD equivalents by a pretty wide margin.
So I can run yesterday's eminently suitable hardware- at 2.5% of the cost.
That's clever, as long as you have an app that scales proportionally across many server nodes!
Dell stuff is a lot cheaper when you A) get a rep and B) buy it at the end of their fiscal quarter :)
I love how the server is on casters. We had some similar servers at one of my old jobs.
What's also remarkable is that you can get all those server features in a 2U formfactor!
Captcha: you just know it's an experiment by Jeff to show that most captcha solutions are overly complicated.
I'm an admin who reads a few developer blogs out of interest.
This post basically reinforced my stereotypical notion that developers should just stick to talking about developing software (and, in your case, usability, I love those posts!).
Things like "even on a single "whatever box" server." would be funny if they *didn't really happen in the wild*.
Running critical apps (and basically, if they're not critical to the core business of "making money", why run them?) on single mock servers that are basically desktops squished to 1U is basically insanity defined, but it happens and the guy who suggested ruining customer trust gets rewarded for "pushing cost down" and the admin who's stuck with a failing piece of metal running a non-certified, non-vendor-supported environment gets the blame.
Additionally, those cheap servers tend to produce noticeably more heat and take more power (except for Dell, whose products I wouldn't touch with a 10' pole anyway if I can choose it), create more "where to get replacement parts 1 year after original purchase" horrors, etc.
What you're pushing here is just a decline in operational culture and ITops understanding presented as progress, as Moore's law is not really news (which you acknowledge).
If there's money to be saved on a large scale by working smarter, it's desktops and support.
Normal companies should once again look at thin clients, and Web 2.0!!!!1! companies should just give their developers the best Workstations that they possibly could, and invest the savings in development time in some decent servers. It's not like cutting cost is all there is to it, because else the most successful companies were the ones that just cease operations (wow! zero cost!)!
Summary: WHY DID YOU MAKE ME FLAIL? ;)
Well, bnerd, I don't agree. I used to write for TechReport, so I have a lot of experience with computer hardware. I think a lot of the traditional server costs ("certified", "vendor supported") are waste, and can be duplicated with off the shelf whitebox parts that are trivially easy to replace. DIY!
*If* your org has a defined IT infrastructure, I'm not proposing that be thrown out and replaced with a random assemblage of ad-hoc servers built from newegg.
However, for small companies starting out on shoestring budgets, a heavily formal IT infrastructure is a giant waste of money.
Jeff, I take issue with the phrase, "starting out," on shoestring budgets. Small firms tend to be on shoestring budgets at all times, whether they're starting out or well into a mature upgrade cycle :)
Jeff, I was flailing more out of professional frustration with being in a bad spot as an admin because of arguments much like yours, than out of disagreement that a sensible IT env can be built out of high quality parts.
It definitely can, and in the grand scheme of things, wonderful things happened on white box boxen.
The problem is basically "quality parts", combined with what Richard says.
I think the disregard for operational issues comes from the plain fact that admins are not as smart as developers (I'm certainly not as smart as the people who develop software as well as I admin), and the resulting idea that it's child play - which is propagated by many developers who teach managers that bad lesson.
The problem there is that I think while coding is a genius-and-hard work profession, adminning is a sweat, sweat more sweat profession; and a lot of time is needed to develop the proper skills and contacts.
Just my 2 cents.
If you have S-ATA drives, then you can swap them out without powering down, unless it is the boot drive.
I just wanted to see how the captcha works since everyone was talking about it