Hacker News Reader

Comment by chalst | Hacker News Reader

"Huge" means different things in different contexts. A compiler optimisation that delivered a 4% increase in speed for a whole class of well-tuned C programs would be regarded as huge by people who follow compiler technology, but would not strike most users of computers as a big deal. I take it "huge" is meant in this sense of extraordinary within a class of potential improvements.

If you can accumulate a series of 4% improvements, then even the computer consumer will start to see the significance.

View on HN

A 4% improvement of the 100m dash world record would have been considered phenomenal, I imagine.

This reminded me of the 4 minute mile. The first 4 minute mile was a 3:59 run by Roger Bannister in 1954. Prior to that, the record was a 4:01 run in 1945.

A 4% improvement on the 4:01 would be a 3:51. A 3:51 wasn't achieved until 1966, about 20 years later.

So 4% can be quite a jump!

Extending your point, 4% more would go down to just under 3:42.... which has not yet happened, over a half century later.

To put that into context, Bolt's 9.58s (current WR, considered phenomenal in and of itself, I think?) is about 1.7% faster than Powell's 9.74s (last pre-Bolt WR).

A 4% improvement on Bolt's 9.58s would be about 9.19s - probably not something we'll see for several decades, if ever.

I wonder if I’ll ever see a Bolt again in my lifetime.

Such dominance and grace.

He jogged over the line with that record too.

I can’t find where the 4% number comes from, but I expect that it comes from a micro benchmark that does little more than futex calls. Reason: if it is from a benchmark that represents real-world use, real-world use must spend at least 4% of its time in futex calls. If that were the case, somebody would have seen that, and work would have been spent on this.

So no, this won’t give you 4% in general.

From the lkml post, it's neither from a real-world case nor from an especially constructed futex benchmark, but from an existing kernel benchmark where it was thought it would be interesting to see how it performed with the new call.

> If that were the case, somebody would have seen that, and work would have been spent on this.

I would think that the linked patch is good evidence that work is being spent on this.

The microbenchmark part is part of the article, no idea why the downvotes here.

Because the benchmark isn't doing what GP days it does.

Reply on news.ycombinator.com