Hacker News Reader

Comment by justadudeama | Hacker News Reader

How do you think the IRS should go about updating? Do you thing something like AWS is stable and safe enough fro something this important?

View on HN

Why not? AWS builds airgapped datacenters for spooky government agencies. The Pentagon wanted to migrate before Oracle stepped in with lobbying. Those private DCs are rock solid.

As a side note, I think that's just amazing. The ultimate IaaS: your own private AWS.

I'm skeptical of the idea of the IRS moving to a contemporary platform of any kind. They have 50 years into the current configuration, with no signs it will be replaced soon. On the time scale of decades, what platforms have proven to be stable and backwards compatible over that time frame? I mention the decades time scale, because I'm not at all confident that the next revision (after the one that hasn't yet happened) of the IRS platform will take anything less than another 50 years, and in that time, almost everything current today will look like a flash in the pan.

I think they'd be better off moving up one step in the abstraction layer with something written in COBOL for the backend, and some kind of service layer whose clients can be updated over time as need allows. If they can de-couple and modularize as as possible, it will make future upgrades more manageable.

In many respects their data workload pales in comparison to what used to be called "big data". Their system is more analogous to airline scheduling in the need for availability, correctness, and longevity. They have to take a very conservative approach. Even if they were spending tens of millions a year on maintenance, that's a better option than the risk of a major downtime during tax season.

AWS already hosts government services:

https://docs.aws.amazon.com/govcloud-us/latest/UserGuide/wha...

Start with really simplifying the tax code, and then do a complete rewrite.

Or wait until the whole mess collapses, and start with really simplifying...

That's what's tough about public service. You have to cope with what is already there.

So how, exactly, would you simplify the tax code? And make sure it brings in the same amount of revenue, and isn't extremely regressive? And make sure it isn't trivially game-able?

Oh something like no tax up to a certain level, 3 or 4 bands from then on, similar with inheritance tax, but a firm no dodging it policy for the very high value estates. Crack down on all the trusts etc. Toss it into a decent simulation and make regional adjustments depending on local income, and actively try to address some of the regional imbalances over time with national funding for things like schools.

It is btw. far easier to game a complex tax system than a simple one. Talk to your current President about that.

At a certain point, the technical debt on this kind of thing is so large the only thing to do is toss it out, and start again.

"Oh something like no tax up to a certain level"

Already have that.

"3 or 4 bands from then on"

Already have that.

"similar with inheritance tax, but a firm no dodging it policy for the very high value estates."

Already have that.

"Toss it into a decent simulation and make regional adjustments depending on local income, and actively try to address some of the regional imbalances over time with national funding for things like schools."

How is that going to make figuring out taxes easier? What if I live in one cost of living area for a while, then move to an area with another cost of living? Now I have to keep track of what income I earned where?

It's good enough for the CIA. Why couldn't it handle a glorified spreadsheet?

Easy -- U.S. citizens don't care about who their government spies on/if the spies get hacked. But the notion of some Russian/North Korean/Chinese/Choose-Your-Favorite-Bad-Guy hacker messing with your money? Unacceptable!

Given that Azure and AWS now have/will have regions certified for National Security (AKA classified) workloads, absolutely. (Disclaimer- I work for a major cloud service provider). But when trying to replace a super high profile, monolithic, contentious government system like this your problems are going to be much deeper than what infrastructure/platform to pick. This is project where, fundamentally, a large part of the organizations board (in other words the further right wing representatives in congress) are basically opposed the entire mission of tax collection. The tax code (which is basically the requirements doc) is 16,845 pages long (although I don't know how much of that is individual income tax related). That's before for all the analytics, reporting, interfaces, and other add on's that the design committees take on.

I can almost assure you that's the MASSIVE requirements tome they dev team is dealing with. They'll probably try to break things up and do things in an agile way. They may even have all the trappings of Scrum up to a Scrum of Scrums organization, but my guess is that the executive teams in charge are going to have a hard time coping with anything besides good old waterfall style implementation and can't fathom releasing the system in small parts (and with the ancient monolith in place it may be very difficult to so anyways.)

On top of all that, this system WILL be the subject of multiple lawsuits, even if it behaves exactly the same as the old system- people love suing about their tax bill. This causes all sorts of hand wringing and worry from the executive team, seemingly crazy refusal to change away "from the way we've always done it", and countless review meetings.

Finally the very existence of the IRS is treated with distaste by large number of members of congress, who will drag any flaws in the system up in front of committee and pillory everyone involved.

Given the project context, it'll be a real achievement to get it updated at all. So long as they are able to break up the monolith, it doesn't really matter what it targets at the start, at least there becomes a hope it can get moved later.

Reply on news.ycombinator.com