Comment by telendram | Hacker News Reader

Comment by telendram | Hacker News Reader

Parent Full thread

This is a deliberately misleading statement.

When compressor A reduces file size by 90%, and compressor B reduces same file by 99%, it's not "9% better". It's 10x times better. Because it can fit 10x more data in the same storage budget.

The rest of the paragraph get exactly the same kind of bias. Take the absolute slowest mode, which is even guarded by special options, and make a general statement based solely on the speed of the slowest advanced setting, implying everything is slow. This is another willful misrepresentation, as zstd is built primarily for speed, and at its default level, compresses not only more than gzip level 9, but also at more than 10x the speed. Not even mentioning decompression speed which is much faster, nor the integrated multithreading mode which can make the difference even more startling since it is not available in default gzip.

In short, this is a troll statement. You have picked a side, and contort representation to correspond to your pre-conceived agenda. At the age of disinformation, when bots and troll farms are deliberately pushing confusion onto unsuspecting readers, we have to call these acts as they are.

> This is a deliberately misleading statement

> the same kind of bias

> another willful misrepresentation

These phrases, as well as the whole last paragraph, contribute no meritorical value to the comment, but are aggresively vocal about assuming bad faith and as such violate HN guidelines. Please don't do that.

I disagree. Lies should be called out as lies, and aren't nearly often enough in the current era of fake news/science/politics.

IMO, attack the message, not the messenger. Pointing out flaws in an argument is good, but assuming malice hurts the discussion. Now we are suddenly talking about the morality of the poster instead of the issue at hand.

Never attribute to malice, that which can be adequately explained by error.

https://en.wikipedia.org/wiki/Hanlon%27s_razor

Well, I did say "about 37% if you consider .gz as baseline", but I agree that it could have been worded better.

The problem is what you are measuring. If you have to compress a 100MB file, and algorithm A/B/C/D compress it to 10/9/1/0.9MB, respectively, then we can argue that "B is 10% better than A" (because the file is 10% smaller) and also "D is 10% better than C" (same).

However, moving from A to B saves us 1MB for this file. Moving from C to D only saves us 0.1MB. So depending on how you look at it, you could also argue that the difference between C and D is really only 0.1% (of the original file size). It may not be what you want (for a particular situation), but it's not wrong.

> at its default level, compresses not only more than gzip level 9, but also at more than 10x the speed

I could also call that a wilful misrepresentation, since gzip level 9 is very slow compared to the default setting and IMO isn't worth the extra execution time.

Also, nobody mentions that gzip is already installed everywhere. zstd is fairly new, and it's not installed by default on any of the machines I use.

I mean, I'm sure there are some exotic screw drivers that work better than Torx. But I can get Torx screws and drivers in every hardware store and everyone already has compatible bits so why not use whatever works?

Never trust any statistics you have not manipulated yourself...

You could have demonstrated the math without your attacks.

> It's 10x times better. Because it can fit 10x more data in the same storage budget.

This makes no sense. It's 9% better because it's 9% better.

The resulting compressed archives of these operations are 10% and 1% of the original file sizes, respectively (10x difference). The percent reduction can’t really be compared linearly.

I share your frustration with misinformation, but this is just human error, which you could be much kinder about correcting, I detected no gzip-pushing "agenda" in that comment, are you sure you're not just projecting?

Reply on news.ycombinator.com