1
100%
master: 100%

Ran 31 Dec 2016 12:55AM UTC

Files 66

Run time 3s

Badge

Embed ▾

Committed 30 Dec 2016 11:28PM UTC coverage: 93.606% (-0.07%) from 93.675%

Job # TERM=dumb

Build Type

push

travis-ci

Committed by

ben-manes

Commit Message

Use simulated annealing to optimize window size

This builds on the simple hill climbing approach. The idea is that the
admission window size can be optimized by making an adjustment, observing
the hit rate change, and adjusting again. Ideally this climbs the hill to
the optimal configuration, up or down based on the feedback. The simple
approach had a lot of jitter and moved too slowly, but showed promise.

Simulated annealing uses a cool down rate to slow the adjustment process
as it nears an optimal value. This also lets us reduce the pivot to take
smaller steps as we zero in on the target. The process halts, thereby
ignoring jitter, until the hit rate changes dramatically (new workload)
or a random restart is triggered.

The adaption method should be further fine tuned. There may still be too
much noise, especially due to resetting the hit rates so frequently. We
should probably use a weighted average across a few hit rate samples to
avoid noise. But overall we see a nice improvement.

Coverage Stats

5387 of 5755 relevant lines covered (93.61%)

0.94 hits per line

ben-manes / caffeine / 1939 / 1
100%
master: 100%

README BADGES
x

Markdown

Textile

RDoc

HTML

Rst

Source Files on job 1939.1 (TERM=dumb)

ben-manes / caffeine / 1939 / 1 100% master: 100%

README BADGES x

Markdown

Textile

RDoc

HTML

Rst

Source Files on job 1939.1 (TERM=dumb)

ben-manes / caffeine / 1939 / 1
100%
master: 100%

README BADGES
x