This is the initial SLIDE paper, this is supposed to be an improvement on top of SLIDE. Having looked at and run the original implementation I'm a lot less skeptical than the average post here. It was quite fast already and should be considered a toy implemenation.
I think the reason you're getting so many questions about the 'toy' statement is that often toy programs are simple in large part because they omit handling complex or difficult cases. I don't think I've ever heard the term applied to a complete problem that is merely implemented naively.
As a result, saying it's a toy implementation is making people think you mean the speedup is a result of simply handling a simple portion of the problem, rather than thinking you mean even a naive implementation is quite fast.
> this original was a toy/PoC and this new paper is much better
This one hundred percent, the original has very few optimizations over a completely naive implementation. It uses MPI and huge pages, that's essentially it.
