Hacker News Reader

> Putting author names but no date at the top seems to be some sort of standard template or convention that every single paper author or institution everywhere uses since forever, since I never see dates at tops of papers.

The year usually appears in the bottom left of the first page, and in the header or footer of subsequent pages.

The first linked PDF [1] says only "14th USENIX Symposium on Operating Systems Design and Implementation" (without the year), but does have the year on an initial cover page. The second [2] says "HotOS ’23, June 22–24, 2023, Providence, RI, USA". The third [3] says "SOSP ’23, October 23–26, 2023, Koblenz, Germany" in the bottom-left of the first page.

However, this information may be missing from an author produced files/preprints, as it may be inserted along with page numbers when the conference proceedings or journal issue gets created. This seems to be the case for the fourth linked PDF [4].

[1]: https://www.usenix.org/system/files/osdi20-yang.pdf

[2]: https://jasony.me/publication/hotos23-qdlp.pdf

[3]: https://dl.acm.org/doi/pdf/10.1145/3600006.3613147

[4]: https://junchengyang.com/publication/nsdi24-SIEVE.pdf

Links like https://junchengyang.com/publication/nsdi24-SIEVE.pdf is the one we can actually read though, relying on a third party to put this critical piece of information seems suboptimal, rather than make the paper self-contained.

I noticed that arxiv at least puts the dates in vertical on the left side of the PDFs there, but it seems that too is something arxiv puts there, not the paper itself.

If it's about subtleties like "what is exact publication date", at least put some date, at least let us know it's from the 2020's or the 2000's, it's not guaranteed to be easy to tell. It should be possible to put a year during which the text is being typed there.

Papers usually use publication date, and this is a preprint - it is to be published in NSDI'24, which will be in 3 months from now.

Usually the final paper PDF, from the conference proceedings, contains the publication information.

Articles are being written about it now, cool that it's in NSDI'24 in 3 months, but that seems almost irrelevant if we're talking about it now.

Wouldn't the target audience of a preprint also benefit from knowing a rough date?

> Well turns out the paper is from 2023 or newer, so that could explain why.

TFA and the paper both wonder why this extremely simple algorithm was not discovered sooner. Both posit that no paper should be prerequisite to applying this modification.

The question in TFA is less about the paper and more about the concept -- is there some reason why this has evaded us? Is it too good to be true?

The article does a better job than the paper digging in to the details IMO.

Asking why humanity didn't invent an invention earlier, given that we had all the prerequisites, is kind of a banal question. Why didn't we invent cars sooner? Why didn't we invent the internet sooner? We had all the prerequisites, or else they wouldn't have been invented. Why didn't we invent them faster?

The answer to all these questions is that having the prerequisites for inventing something doesn't mean we instantly invent that thing, a la Sid Meier's Civilization. The expectation that we would is actually what's kind of odd here. Humans doesn't even all share the same info, much less automatically know all the possibilities of all combinations of all humanity's knowledge.

Another factor might be that Yet Another Eviction Algorithm isn't necessarily one of humanity's top focuses for inventions at the moment. We could ask any of us here on HN, for example, why we didn't think of it, and sooner. The answer is likely that we were all working on something else we found more important than YAEA.

That banal question is in the original paper. The paper spends significant time explaining that this because this is such a simple modification they find it difficult to believe it's been overlooked until now.

The article's question "why aren't we SIEVE-ing?" follows this theme.

No one is asking the equivalent of "why didn't we invent cars sooner?" here.

I don't doubt that the banal question in question, is in the original paper. My point was how banal the question was, and how easily answered it is, not where it was.

Many inventions in humanity were "simple modifications" of other things. Why didn't we invent them sooner? The answer is in the post you replied to: because it's totally normal for "obvious" inventions to not be invented for a long time, even when somebody thinks, after the fact, that it's a "simple modification". Indeed, what percentage of humanity is currently searching for Yet Another Eviction Algorithm, and what percentage of that percentage is looking in this direction, instead of one of the hundreds of other directions?

So yes, the question is indeed equivalent to "why didn't we invent [X] sooner", replacing X with any invention that at least 1 person out of billions in the world finds "obvious" or a "simple modification".

Not everything is an invention.

Do you have any routines that you've refined over the years? Is every step you take in a routine something you were taught or learned elsewhere? Or did you "invent" part of it yourself?

I know how to make tea or coffee. Should someone take the credit for the fact that I know to start boiling water first, while I get everything else together, because this results in the most efficient use of my time?

My perception of the paper is this analogy isn't too far off. It's not an invention, it's a discovery.

Perhaps not everything is an invention, but I think everything we're talking about here is, including algorithms. In any case, that's just semantics: you can call it a "discovery" if you feel better about it, perhaps even run s/invention/discovery/g on my post.

The points therein still stand: [inventions/discoveries] which at least 1 person in the world thinks (after the fact) are obvious and just a "small modification" of other [inventions/discoveries], usually go [uninvented/undiscovered] for quite some time.

The paper and TFA asks, "why?" But the better question is, "why wouldn't they?" There's no reason to think they would be be immediately [invented/discovered] as soon as all the prerequisites are [invented/discovered] (again, even if someone thinks they're similar), for the reasons described in my above posts.

Or: why did we invent X so early, far earlier in history than seems logical given the "tech tree" view we get from sim games?

The fax machine and the electric car come to mind.

Electrical car makes sense to me, when cars were new all kinds of propulsion methods were attempted, eventually one dominant technology gained traction.

The electric printing telegraph in 1848 though, that's amazing!

I just use the publication date.

Just looking at NSDI ‘23, I can see the first published paper in the Spring: https://www.usenix.org/conference/nsdi23/presentation/liu-ti...

Looking at the BibTex metadata can give you a better idea of how recent/relevant the paper is:

year = {2023}, … month = apr

But since this is not yet published (to appear in 2024 of NSDI), I use a similar method you describe. Most of the time, can also infer from the paper itself (sometimes in the abstract or the discussion)

I never noticed this before, but you're right, it is very weird!

> or newer, so that could explain why.

The title is more raising the question of why an algorithm that performs this decently and is so simply doesn't seem to have been invented earlier.

I was thinking this same thing the other day when reading a paper. Seems like a critical piece of information that is often left out.