Slashdot Mirror
Welcome to the New Slashdot Chicago Cluster
Thanks to everyone who tested on Friday, as well as to all of SourceForge's netops crew, our corporate overlords at SourceForge for paying the bill, and of course all the engineers on Slashteam- Jamie McCarthy, Tim Vroom, Chris Nandor, Chris Brown, and Scott Collins, we are now running on the new iron in a cage in Chicago. We'll run a story in a few days about the ridiculously overpowered new hardware we have now, but now is the part of sprockets where we dance.
So... what is going to happen to the old hardware? Are you going to going to scrap them? Hand them over to sourceforge? Sell them to another company?
Or auction them off to your readership for charity?
So now that SourceForge has upgraded Slashdot's fancy-pants hardware, when are they actually going to make sourceforge.net not suck so much?
Just kidding. I love you guys.
But seriously, sourceforge.net sucks balls.
Do you guys have a description of the migration steps hidden away in a journal somewhere?
Appreciatively,
Seth
$5 / month hosted VPS on linux = awesome!
For I am but a head in a rusty metal box.
In the computer world, there is no such thing as 'ridiculously overpowered'.
:P lol
Can your servers run Crysis on max settings?
Well, the search feature isn't working.
So everything's back to normal.
Blank until
However, my employer would love to find your new cluster and beat it with a tire iron.
John
This is absolutely news.
Finally, Slashdot has the highly redundant replicated infrastructure worthy of their highly redundant, replicated article topics.
There are no karma whores, only moderation johns
Chicago is the center of the universe?
Actually, my work is considering where to move it's servers. It involves a HUGE amount of fact finding. Chicago is one of the places they want stuff, but that's for customer reasons, not for "center of the universe" reasons. Our locations are chosen based on current customer usage, and statistical information I gathered at previous jobs. When you have 8 million users/day from around the world, those demographics stick in your head real well.
In my research, I found the best places to be are.
New York City. 111 8th ave, 60 Hudson, or 25 Broadway. The selection would be based on provider interconnects and availability. Some providers service all three locations with their own private interconnects, so it really doesn't matter.
Los Angeles. One Wilshire, or one of a few select locations nearby, again with private interconnects to One Wilshire.
Miami. Near or at 1 NE First St.
Chicago. Near or at 427 S La Salle St
The runners up are:
Chicago
San Jose
Amsterdam
Frankfurt
London
Paris
Tokyo/Osaka
In time, I'd like to have equipment in all of those locations. Or we can go the Akamai route, and put stuff anywhere there's a rack.
For just about any provider of English based contact, the rankings of customer location by major geographic area are:
North East United States
South East United States
Europe
Western United States
Obviously that would be skewed for the content. For example, a Japanese speaking site, with local interest content would be best placed near JPNAP in Tokyo or Osaka. Likewise, a Russian site with say daily weather reports of Siberia would probably want to be in Chelyabinsk, Russia, and you probably want to use Rostelecom.
I noticed that Slashdot is now using Savvis. They were offering an amazingly cheap deal on bandwidth recently. I wasn't actively pursuing the bandwidth side, I was looking for the physical location side where my providers of choice would be. I'd be willing to bet they're in the Telegraph building. I'm curious now to who's suite they're in.
Serious? Seriousness is well above my pay grade.
Screw that. Tag it touchmymonkey. Do it, you know you want to!
Having worked for many large websites that have a boat load of visitors every day I think I can give a little insight. (As well as having worked for many web hosting datacenters)
More than likely they have a good deal on bandwdith, assume they use 500Mbps a month, a traceroute from seattle shows they're using at least savvis as one provider. We can safely say they're probably paying no more than $18.00USD per month per Mbps. These are prices a small time carrier might get, they may be paying more and who knows how many other providers they're paying for or if they just buy a blend from their colo center.
Lets put that price at: $9000/month so far
Then you factor in the floor usage, normally you can get a full rack for about 2000 with 40a. I'm willing to bet they use at least 60a per rack, add in another 500. Plus they say they have a cage, they charge extra per square foot, a guess would put the cage at an extra 1000 per month.
So I'd say about 3 racks with a cage would cost around: $8500/month
So far the total is about $17,500.
Now if they staffed their own people and didn't have any outside monitoring or anything of that nature that might be the total cost. In reality they probably have a contract with their provider for one site maintenance, 24/7 on site support, hardware replacement and the likes.
At my current place of employ we pay 30,000/mo a month for that kind of service, I think we're a bit above what one would normally pay but we have a pretty high up time SLA.
Add another 10,000 a month for maintenance/support/supply contracts.
Grand total I'd say is about $27,000/month USD. It might be higher or lower depending on their deal with their providers but normally for a standard colo deal it's around that price.
I've seen sites pay out well above that (100,000+) for colocation and have an awesome return.
Greetings Germanic Dancers,
:)
Could you please add an IPv6 VIP to slashdot now that you've got this move out of the way? I mean, it's 2008 already
Have you ran any stats on your dns logs to see what percentage are requests for quad-As?
If you're nervous about suddenly blackholing because of misconfigured remote sites, perhaps you could add an ipv6beta.slashdot.org site à la ipv6.google.com? Or, I read of a long-running test a website had been running where a third of clients were served a one-pixel image from a hostname with a single AAAA record, another third a dual record, and finally a single A record to test against reachability problems.
So, I'm sure you're all smart and working on it and I'll just have to keep patiently waiting, but I'll be so pleased when your v6 integration matches undeadly.org.
Its not the lag that causes problems, its the rotating the bits so that they're the right side up that causes the delay.
rewriting history since 2109