Hacker News Reader

Comment by blt | Hacker News Reader

SIEVE isn't just playing the part of a cache eviction algorithm; it's stepping up as a cache design superstar. Think of it like giving a fresh spin to classics.

Is this how academics write about their own work in blog posts now?

The result seems really strong. It's sad the author feels this kind of hype is necessary.

View on HN

Recent hype cache algorithms are characterized by cherry-picking hit rates in content caching such as CDNs. Common cache algorithms such as ARC, LIRS, and TinyLFU have made consistent and common comparisons across a variety of disk access patterns, but hype cache algorithms boast hit/miss rates for network accesses without these comparisons. Hype cache algorithms have only been validated in a very narrow domain, and their general superiority has not been validated at all.

Just to clarify, this means hype cache algorithms are basically not replaceable with common cache algorithms such as LRU or ARC.

I probably should have written the opposite. English is unfamiliar.

I haven't read the original article yet but your excerpt here reminds me strongly of what ChatGPT writes when you ask it to rewrite technical prose in an approachable, non-technical "for dummies" style.

As in, I asked ChatGPT to rewrite some text with metaphors and it kept giving me answers that included the phrase "Think of it like...." And whatever the metaphor it came up with was. It also often used words like "superstar" far more than I would.

I agree that the hyped-up language is off-putting.

It is however not at all typical for academic publications and blog posts.

In the few times I have seen it before, the author was always a PhD student and a non-native speaker (as it happens from China) and this case is no exception. They were also all smart people, so I am optimistic that with time their writing style will become more attractive.

That caught my attention too. It mixes the level of communication jumping from highly technical to kindergarten when you have not a cognitive seat belt for that.

Also why? The only value that comes out of that comment in the middle of the technical discussion is vanity and nothing else. It could have been more useful if it would be creating a metaphor that approximates understanding. That would be forgivable. But vanity isn't.

I believe there is also a paper [1] if you're looking for more details or a more academic description.

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

Impressive how few code changes they managed to convert an existing LRU library into SIEVE.

https://github.com/cacheMon/lru-rs/commit/1c45ef7fe85f4a4395...

I presume you have not experienced being an academic in recent years and having to deal with the expectation to sell your research on "Scientific Outreach Twitter". Now, technically neither have I, but I've worked alongside scientists for a couple of years who do.

Yes, it's kind of sad, but I feel more empathy for situation the author is in that leads to it. The actual papers themselves seem pretty solid from a surface reading as a non-expert, at least in terms of having clear explanations and easily verifiable results.

> The actual papers themselves seem pretty solid from a surface reading as a non-expert, at least in terms of having clear explanations and easily verifiable results.

If you only read one paper in an unfamiliar field, you will not know if it lacks the necessary analysis.

This isn't exactly a paper on room temperature superconductors...

I read plenty of compsci papers and whitepapers, I just don't have to work with this specific niche of web caching. I doubt the methodology for benchmarking is so radically different from other computer backgrounds that my ability to detect fudging in it, or other structural issues with them would be compromised.

And on top of that, in this case both the code and data sets are available and the papers went through peer review.

> This isn't exactly a paper on room temperature superconductors...

Same in every field.

> I read plenty of compsci papers and whitepapers, I just don't have to work with this specific niche of web caching.

You clearly have no expertise in this field. Over.

Ah, you didn't come here to have a discussion, but to "win" a "debate". Sure, whatever, you win

I also feel empathy. My "sad" is more about the state of the community than the author's own attitude.

But still, even in the current climate, I think this is going too far. The whole paragraph after the part I quoted is just as bad.

I didn't even notice the style. I first read the verbal description of the algorithm, got angry that it's incomplete and I lack context of what this is trying to achieve. Then I skipped to reading Python code. Then got back to verbal desc to confirm. Concluded it's not that bad but still incomplete. Then I looked at the performance graphs and concluded I might use it if I ever need something like that.

Why do you guys read prose at all when it's obviously just a filler?

Reply on news.ycombinator.com