5 ms·

A contract the size of JEDI can (and probably is/will) build in that kind of independence as a requirement. Which isn't to say "Amazon must build and maintain a

A contract the size of JEDI can (and probably is/will) build in that kind of independence as a requirement. Which isn't to say "Amazon must build and maintain an OpenStack implementation for DoD", but rather, that Amazon must turn over all APIs and source code used to build the implementation. Obviously, source code and APIs isn't enough to keep everything running - there still needs to be operational expertise within organizational memory - but the DoD would be empowered to start to take over operations and plan a migration to some other company maintaining the cloud infrastructure, if it were really untenable to renew the contract in a decade. Put it this way: there's an old aphorism about how owing the bank $1 million is your personal problem, owning the bank $1 billion is your bank's problem, and owing the bank $1 trillion is your country's problem. There would be a parallel for cloud infrastructure: if you have a $1 million cloud bill then you have a (vendor lock-in) problem; if you have a $500 million cloud bill then your vendor has a problem (because you represent a huge chunk of business that can evaporate if you move to on-prem); if you have a $1 billion cloud bill then the wider market would have a problem (because of the reverberating shock effects if your migration to cheaper infrastructure is unsuccessful). It's the textbook definition of something which is too big to fail.

danenania 7y ago
"the DoD would be empowered to start to take over operations and plan a migration to some other company maintaining the cloud infrastructure" That's still a massive project though that I presume would only be undertaken in extremely dire circumstances, so Amazon only has to do the bare minimum to make sure things don't reach that point. I'm thinking more along the lines of switching off the primary cloud to a "follower" cloud in a few clicks. I realize that's not at all easy, but I think it should certainly be possible with these kinds of budgets to build in cloud-agnosticism from the ground up. It's hard but it's not that hard--it's just the facade pattern, which is quite commonly applied in other cases like OSes/databases that are arguably just as complex.
- ethbro 7y ago
  This is how you end up running on an extremely limited cloud subset, decades behind current offerings. Do you want seamless ability to transfer work to another cloud, or do you want the latest managed services and features? Pick one.
  - YawningAngel 7y ago
    When you're spending ten billion, it seems reasonable to run your own data centres to me.
    - ethbro 7y ago
      On government salary-capped architects and engineers? DoD is never going to be able to pay what any of the major companies can for top-tier talent. And there are only so many talented people willing to work below market rate for other reasons. (Not at all a slight to those engineers doing government work. Salutes to all of you!)
      - ryacko 7y ago
        DoD pays for top tier talent when it is a priority. Thousands work as contractors for JPL or the Battelle Institute. The government is currently structure so that only cutting edge research is a priority, not the more humdrum stuff like day to day operations.
      - Supermancho 7y ago
        > On government salary-capped architects and engineers? You mean for the design? Yes. For the actual actual day-to-day operational duties, it's servicemen that perform like any other military post.
      - YawningAngel 7y ago
        Obviously there are lots of ways you can vitiate your ability to do this by being pathologically dysfunctional, all I'm saying is that the proposition isn't a priori unreasonable. Your objection seems flawed to me in that it ignores the fact that goons can screw up just as badly on AWS as they can in a data centre.

danenania 7y ago

"the DoD would be empowered to start to take over operations and plan a migration to some other company maintaining the cloud infrastructure" That's still a massive project though that I presume would only be undertaken in extremely dire circumstances, so Amazon only has to do the bare minimum to make sure things don't reach that point. I'm thinking more along the lines of switching off the primary cloud to a "follower" cloud in a few clicks. I realize that's not at all easy, but I think it should certainly be possible with these kinds of budgets to build in cloud-agnosticism from the ground up. It's hard but it's not that hard--it's just the facade pattern, which is quite commonly applied in other cases like OSes/databases that are arguably just as complex.

ethbro 7y ago
This is how you end up running on an extremely limited cloud subset, decades behind current offerings. Do you want seamless ability to transfer work to another cloud, or do you want the latest managed services and features? Pick one.
- YawningAngel 7y ago
  When you're spending ten billion, it seems reasonable to run your own data centres to me.
  - ethbro 7y ago
    On government salary-capped architects and engineers? DoD is never going to be able to pay what any of the major companies can for top-tier talent. And there are only so many talented people willing to work below market rate for other reasons. (Not at all a slight to those engineers doing government work. Salutes to all of you!)
    - ryacko 7y ago
      DoD pays for top tier talent when it is a priority. Thousands work as contractors for JPL or the Battelle Institute. The government is currently structure so that only cutting edge research is a priority, not the more humdrum stuff like day to day operations.
    - Supermancho 7y ago
      > On government salary-capped architects and engineers? You mean for the design? Yes. For the actual actual day-to-day operational duties, it's servicemen that perform like any other military post.
    - YawningAngel 7y ago
      Obviously there are lots of ways you can vitiate your ability to do this by being pathologically dysfunctional, all I'm saying is that the proposition isn't a priori unreasonable. Your objection seems flawed to me in that it ignores the fact that goons can screw up just as badly on AWS as they can in a data centre.