Profile avatar
omniledger.io
I have built the first async, consistent data-platform https://omniledger.io/ I also write about distributed systems https://medium.com/@andrasgerlits
277 posts 128 followers 200 following
Regular Contributor
Active Commenter
Posts 18 Comments 32

Cloud-providers are selling both the disease (stateful microservices) and the cure (strong consistency). I don't think they are blind to this fact.

submitted 12 days ago • 0 comments

The moat the market priced in for hyperscalers is their ability to precisely control a physical network and through this, latency. Literal trillions hang on the idea that it's impossible to build a comparable network via the public internet. That's quite a gamble.

submitted 13 days ago • 0 comments

I know shooting at random political tweets is below par, but more than 5k likes on someone saying that Musk's team is "the people" and "not the government", to mean "the good guys" shows how words don't mean anything anymore.

submitted 17 days ago • 1 comment

Does anyone know of a project where they combined temporal logic and property-based testing besides Quickstrom? I can't be the only one who needs something like this.

submitted 17 days ago • 0 comments

As a kid, I could never understand why people mistook drilling on tests for intelligence. Now that I have kids I understand it even less. The more I have to make my daughter participate in the process the more convinced I am that this is more about obedience than anything else.

submitted 21 days ago • 0 comments

My god. I've seen countless attempts at rewriting legacy systems, but the mainframes either managed to beat them back or the org was left with a complete mess (usually of the microservice-hell variety) that made headline news and caused days of outages.

submitted 22 days ago • 0 comments

Based on the SQL-spec, why do we need to use locks at all (in an MVCC DB), when we do REPEATABLE READ or READ COMMITTED? Why can't we just expose changes at commit with a future timestamp and read with a past timestamp? Is this like a "vestigial organ" thing?

submitted 24 days ago • 1 comment

Due to my family background, I've been hearing that AGI is imminent for about 35 years now. I've seen four generations of engineers first being excited about the "endless possibilities" to become better pattern-recognition. But sure, maybe this wave will be the one.

submitted 30 days ago • 0 comments

I've been thinking about this for decades and could never understand how entropy can be considered a universal property of any system if entropy is an observer-defined property. Apparently, this has also been bugging Wolfram. www.youtube.com/watch?v=0YRl...

submitted 34 days ago • 1 comment

Paraphrasing a founder in a private conversation: "He's not incoherent because he doesn't understand what he's talking about. He sounds incoherent because so many of his clients are irrational when it comes to AI budgets and he's discussing their buying patterns." What strange times for IT.

submitted 35 days ago • 0 comments

I've been losing my job to an AI for what, 3 years now?

submitted 41 days ago • 0 comments

We're brute-forcing our way to self-driving cars, to where we can be reasonably sure that we know at what point a human needs to take over. LLMs will be the same. Trouble is, it's much harder to find these points in everyday life.

submitted 51 days ago • 0 comments

There are exactly two things in computing: Semantics and Timelines. This is the reason programming languages and databases are such great abstractions that they became standards. Our problems in distributed systems started when we tried to blur the lines between these two.

submitted 52 days ago • 0 comments

What I'm implying with this thread is that since PCC is implemented on top of OCC, all the attributes of the OCC model become attributes of the PCC transactions. In other words: the best possible OCC model is also the best possible PCC model

submitted 52 days ago • 0 comments

I'm not sure if the "positivity score" on Twitter is mass delusion or not, but if it isn't, I think I just found the first real-life use-case for LLMs: mass surveillance!

submitted 53 days ago • 0 comments

How long can this "you just wait for it" continue until people accept that LLMs won't grow much beyond what they are now? I sure hope it's soon. How is this not a "nerd trap" VCs have been lecturing me about?

submitted 54 days ago • 0 comments

The reason we find it so hard to deploy new code to production is because we expect the kind of consistency from our code with its data (and other pieces of deployed code) that we rarely provide to data alone. If all code is ordered with data, rollouts become transactional.

submitted 55 days ago • 1 comment

I think "observer" and "stateful" are actually the same concepts.

submitted 55 days ago • 1 comment