5 ms·
When the initial JEDI contract was announced and having used AWS, GCP, and Azure I imagined a few scenarios that might come to fruition. Considering the scale a
by ZeroCool2u 6y ago
When the initial JEDI contract was announced and having used AWS, GCP, and Azure I imagined a few scenarios that might come to fruition. Considering the scale and complexity of a large 'enterprise' org AD deployment, I found it hard to believe that Microsoft would come away with nothing.
The most realistic outcome seemed like the DoD would rely on AWS for most infrastructure simply because the people at the top wouldn't know better and regulatory capture. I figured they would, almost begrudgingly, bring Azure into the mix, simply because Microsoft makes it nearly impossible these days to retain a completely on prem AD deployment and how in the hell is a bureaucratic monstrosity like the DoD going to disentangle itself from Active Directory?
Another scenario is that if whomever is in charge of making the final decision is a bit more forward thinking and has some vision, they'll go all in on a multi-cloud strategy for maximum redundancy and resiliency and we might see GCP get thrown in the mix. I think it's really hard to argue against this strategy for an org as important as the DoD frankly. Some people often make the argument that starting with one cloud platform and incrementally moving beyond that is the way to go, but I strongly disagree with this. A true multi-cloud strategy has to start by incorporating multiple platforms from the beginning or else you're extremely likely to get locked into a single platform due to organizational inertia. Not to mention the cost benefits. Locking yourself into a single vendor basically removes all your leverage to take your business and go elsewhere. You really can't bluff your way out of that one either, because what are you going to do? Go redeploy all your applications that were built with proprietary platform specific tools on a _different platform_? Nope.
I think the most unlikely, but also most interesting outcome would have been the DoD goes all in on GCP and the chromebook style of computing, basically tearing everything down and starting from scratch. If ever there was an opportunity for that, this would be it, though there are many arguments against that. I think there's a scenario in which something similar might have been possible for AWS, but in reality it doesn't seem possible, because AWS doesn't really have a competitor to G-Suite / Office365. I know they have WorkDocs, but that's not exactly the same thing.
Having used all three, including Azure, myself though, I really didn't believe a Microsoft only scenario would come about. Among other reasons, and to be fair its been a while, but last I remember Azure's annual downtown far exceeds the other two platforms, which seems like a pretty critical point when considering the systems running on here may hold soldiers lives in the balance.
Anyways, I do hope we see a more reasonable scenario come about eventually. Any single vendor locking in the DoD would be a disaster for national security, soldiers lives, and frankly the federal budget.
- busterarm 6y agoI very recently have built a multi-cloud empire of machines, effectively, and from my experience it is best to build for each cloud one at a time. The difference here is you have the day 1 assumption that you will be on all clouds and you don't lock yourself into vendor-specific features and make sure your platform can run anywhere with no functional differences to your users.
- ZeroCool2u 6y agoThat's fair and totally makes sense from the perspective of someone that is able to enforce that assumption. In my experience in large orgs that condition is very difficult to enforce and that's when organizational inertia kicks in. But your method is objectively better if you can prevent people from doing dumb stuff while you keep working.
- busterarm 6y agoWe handled it by spinning up everything via terraform and forcing dependency on those modules & resource bundles. Separate sets for each cloud provider and it all gets code reviewed by one team. Essentially we built a platform on top of each cloud. The difficulty there is groking the differences in network topology and access control and building something functionally equivalent everywhere.