From: SASADA Koichi <ko1@...>
Date: 2014-01-06T16:06:34+09:00
Subject: [ruby-core:59577] Re: [ruby-trunk - Bug #9262] global_method_cache should be configurable or grow automatically

Hi,

Now, method cache technique is important and we need more measurement
and techniques.

##

From Ruby 2.0, we use inline/global method cache aggressively. So
performance impact on method cache miss has huge impact (compare with
Ruby 1.8, 1.9), that guys already show.

funny_falcon's approach is very impressive to solve this problem.
However, now I'm not sure how it increase memory imapct.

tmm1 measured the memory impact on
https://bugs.ruby-lang.org/issues/9262#change-43840 . This survery is
very impressive. I think it is better we measure other cases. For
example, if some classes calls many methods at onece, it will memory
issue. I think the simple limitation (cap) approach with funny_falcon's
patch works fine.

##

New years holiday (Japanese take holidays in new years week), I'm
thinking about this issue. Some ideas are available.

(1) inline cache
  (1-1) polymorphic inline cache
  (1-2) per class inline cache other than call site
  (1-3) per-method cache invalidation
(2) global cache
  (2-1) per class method cache w/ cap

Now the above ideas are not implemented/verified. And huge effort is
needed (because we need to change the method entry data structure).
Before the try, I need to know the why and how method cache is missed.

##

Basically I don't against to introduce and backport some proposed
patches (w/ measurement, if we can). In my opinion, simple variable
global cache entry size approach will fine for backport.

And also I try above ideas for Ruby 2.2. Current patches are good
starting point, I think.

-- 
// SASADA Koichi at atdot dot net