Slashdot Mirror
What Kind of Data Center Can You Build With $500M?
coondoggie writes "So, if the government gave your company $500 million to spend on building a new data center what would you buy and how would you build it? Well, the Social Security Administration is about to find out. As part of the stimulus bill, or the American Recovery and Reinvestment Act of 2009, the SSA got the tidy little sum to replace its National Computer Center. The SSA in fact says it will need closer to $800 million to fund a new IT infrastructure, including the new data center — the physical building, power and cooling infrastructure, IT hardware, and systems applications. (This is addition to a $72 million backup facility currently under construction in Durham, North Carolina)."
I will start with the assumption that this data center must be non-homogeneous. Get an assessment of all the projects that are using the current system you're going to replace (you know, the one with 36 million lines of COBOL code?). Because the number one priority of the customer (other projects) is going to be the lengthy transition from that to current technology. Prepare yourselves for this: Some of the projects aren't going to have any funding to do jackshit. Which means that the awesome spaghetti coded current system that's held together with COBOL duct tape needs to remain intact in some form. Not ideal situation but an uncomfortable truth. I'm thinking you would want to set aside 10% or $50 million or so for this (just throwing out a figure).
My work here is dung.
Could you approach Google and ask them to license their ideas on server and data 'pod' technology for your sharded databases? I'm not saying build the whole thing like this but with $500 million, you could probably have a large section to search and sharded databases that mimics Google. I don't think there's anything wrong with following the leader in that department. This probably isn't the best solution for relational databases so I would think another architecture would be in place for your MySQL and Postgres traditional database layouts. And that would be just huge centralized servers running virtualized instances of Linux with MySQL or Postgres.
My work here is dung.
Putting the operation in a location that is cost effective would make the taxpayers very happy. The DC area is too expensive. Maybe an old missile facility in Wyoming or the Dakotas.
This is a really really bad place to ask how to spend $500000000....
...but I'll bet you $500 million that: 1) It won't be nearly enough money; 2) It will be obsolete before it is finished; 3) It won't be finished before Social Security runs completely out of money (which will coincide nicely with my scheduled retirement); 4) [Fill in the blank]
or they could contract it out to google.
someone who knows how to manage large data centers correctly.
Dislaimer: I have worked for a contractor to the Air Force, and I have some insight as to government bidding, contracting, and results.
Well, if I were responsible for results, I would get requirements from staff, send out RFPs, hire the best people, manage the project, and deliver on time and on budget a state-of-the-art data center.
If I worked for the government, I would do the following:
- Find a company I would like to work for as a six-figure lobbyist, and hire them without regard to experience or practicality. I will have personal contacts with the CEO, and if I don't, I soon will.
- Get my "requirements" from that company, and have them provide the solution they specialize in without looking at my environment.
- I would not supervise them. I am too important for that.
- I would ask for more money as the project spirals out of control. The government would give it to me.
- The project would drag on for a decade, would never finish, and would ultimately get scrapped. I wouldn't care, because I now work for the vendor, lobbying the government for more projects. I get my own private limo and driver, and I don't have to declare it on my taxes, unless I want a very visible government job again.
Someone might raise a fuss in the public about this, but all that proves is that the government need more money to fix it.
P.S. The contractor I worked for beat out a lower-bidding, "women-owned", "development-zoned", and much more local company. By any government calculus, the local company should have won the bid hands-down. But there were, shall we say, non-written reasons the local company lost and the gigantic out-of-state, double-the-bid, next door to DC company won.
Because they're not in my Congressional district?
There are tons of already built data centers that have been abandoned. They already have much of the infrastructure required and just need some TLC. Using these already built data centers will allow one to have more than one data center, very important for disaster recovery, multi homing, and impressing your customers with your size. I would avoid building a new building from the ground up. Its a cost that will add no value considering the number of buildings available that can be repurposed.
Make strong investments to make sure you can meet you power requires and can readily add more power in a reasonable amount of time. This includes battery piles, back up generators, fuel storage, and maybe even solar panels. Investigate new energy efficient cooling.
It's also important that your data center have easy access to multiple carriers. Look for buildings that used to be ISP central in the late 90s and early 00's. Often these buildings will also already have supporting infrastructure.
DO NOT BY INTO HYPE. Not everything needs to be cat6e or fiber. Use these were required and use cheaper technologies that have specs that will meet the requirement.
Invest services that will add value to your customers. DNS, backups, router maint, firewall sevices, remote hands, terminal services, etc... Dont be afraid to sell your customer shelves, servers, etc... But for god sake, give them screws if they ask. This simple gesture goes a long way to make you not look like small time asshats.
One of the biggest investments you can make is in a person who has real experience in the area and has the ability to get things done. Without someone who understands power, cooling, how to terminate various connections, telco, racks, project management, etc. your project is doomed to failure. Investing in a NOC that isnt full of monkeys will also be a great benefit. (keep them engaged with training and give them a sense of worth and you will create a team loyal to you. Abuse them and they will talk shit to anyone who listens). And your sales engine has to be stocked with people who can sell the services you are selling and can answer basic questions about those services. It might be a good idea to prove that you can sell these services before even breaking ground on the data center to begin with.
Don't over invest and over build. Plan for the future, and use profits to build the next stage. Look at what is coming down the sales pipe and try to predict when you need to add on. Buy customer cabinets and wire them only after the sale has been made.
This could be a great time to invest in experimental technologies for cooling, or to avoid cooling, solar power, etc. It all depends on what the building you find can and will accommodate. But its the bottom line you must always consider. Lots of dotbombs had grand ideas and good intentions and ended up just pissing away all their investors money. Dont dech your NOC out like it was the helm of the enterprise. Dont have large screen TVs and projectors displaying data that the NOC gets alerts on their work stations for.
Oh, and seriously, dont be a jerk about letting your customers use the bathroom, vending machines, and whatever. And have some comfortable couches in the vending area that support rest and work. Some of your customers are going to pull all nighters and there is nothing worse than having to sleep on the floor in a puddle of your own waste chewing on a pizza crust found in the trash.
Have real test gear on hand. If you cannot test a t1, ds3, dsl, pots, ethernet/fiber/coax, throughput, etc you will cause your employees and customers a lot of grief. I'd also put some money aside for a fiber fusion splicer, but dont buy it until you need it.
Data centers can be extremely profitable. I know of one company who I believe had their data centers initial investment paid for with just a few customers in 4 months, after that they began to see profit. Those times might be a bit off, but not by much. (The data center in question I am talking about is in cleveland/garfield. Another option might be to partner with a successful company in another area.)