Re: [Cake] [RFC PATCH 2/3] Add cake related includes and source files

[Dave Taht <dave@taht.net>]

Jonathan Morton <chromatix99@gmail.com> writes:

> On 17/11/17 21:52, Dave Taht wrote:
>>> +#define REC_INV_SQRT_CACHE (16)
>
>> Never been proven that this cache either improves cpu utiliziation, nor
>> that the added accuracy has any effect. I'd like to be able to
>> completely compile out the cache.
>
> CPU utilisation is debatable.  It could go either way, actually, compared to a
> *single* unconditional Newton step.  The cache is probably faster than
> *multiple* Newton steps which would be required to guarantee the same accuracy.
>
> However, a few minutes' work with a spreadsheet is sufficient to show a *big*
> difference in timing between the accurate values placed in the cache and the
> approximations given by a single Newton step for small count values,
> particularly when stepping from count 1 to 2 or vice versa.  Above 16, a single
> Newton step gives sufficient accuracy in either direction, given the precision
> stored.

> The magnitude of the error when stepping up to count 2 is such as to give the
> value that *should* have been produced at count 4.

I am going to add this as a comment.

> Since you are presently
> concerned elsewhere with Cobalt's supposedly "aggressive" behaviour, removing
> the cache would definitely make that worse.

I would like to work towards a stable version that can be benchmarked at
long rtts. The network namespace stuff is helping here to simulate that.