1739
100%

Ran 05 May 2016 06:12AM UTC

Jobs 1

Files 65

Run time 3s

Badge

Embed ▾

pending completion

Build # 1739

Build Type

push

travis-ci

Committed by

ben-manes

Commit Message

Further proving out Adaptive W-TinyLFU is awesome

The first attempt showed that we should adjust when multiple rejections
are detected. It grew the window in recency traces and kept it small in
frequency traces. But, as a bloom filter it didn't properly adapt when
multiple types were run in sequence. It favored a large window.

The next attempt showed that if we should constrain the growth to when
we also detect another accesses. This quickly increased and decreased
the window based on the changing worklaod. However the quick hack used
TinyLFU's frequency (10x max size sample period). It would be fine for
small caches but the recency signal would be less relevant for larger
ones due to the long sample time.

This version uses only the maximum size sample period for both signals.
This was done by hacking another CountMin Sketch variant, but perhaps
two bloom filters would have been better. Deciding on the right sketch
is a future task, as is making this code less horrible. The result was
a further nudge at improving the hit rate with good adaption. I think
we found a winner!

Coverage Stats

5318 of 5631 relevant lines covered (94.44%)

0.94 hits per line

Jobs

ID	Job ID	Ran	Files	Coverage
1	1739.1 (TERM=dumb)	05 May 2016 06:12AM UTC	0	94.44	Travis Job 1739.1

Source Files on build 1739

Detailed source file information is not available for this build.

ben-manes / caffeine / 1739
100%

README BADGES
x

Markdown

Textile

RDoc

HTML

Rst

Jobs

Source Files on build 1739

ben-manes / caffeine / 1739 100%

README BADGES x

Markdown

Textile

RDoc

HTML

Rst

Jobs

Source Files on build 1739

ben-manes / caffeine / 1739
100%

README BADGES
x