1
100%
master: 100%

Ran 02 May 2016 08:15AM UTC

Files 65

Run time 3s

Badge

Embed ▾

Committed 02 May 2016 07:13AM UTC coverage: 94.495%. Remained the same

Job # TERM=dumb

Build Type

push

travis-ci

Committed by

ben-manes

Commit Message

An early prototype of an adaptive W-TinyLFU policy

Experiments have shown that in many workloads TinyLFU is an excellent
filter. However, in some workloads where recency plays a larger role
the frequency filter is too strict. We then devised an admission window
(W-TinyLFU) to capture sparse bursts. In many workloads a 1% window was
satisfactory, by not degrading frequency workloads and helping recency
workloads. However, finding the optimal window size is workload
specific.

Can we adjust the window size based on the workload? If we detect that
TinyLFU mispredicted and the new arrival was rejected twice within a
short period, then we increase the window size. If we see that the this
is not the case, then we decrease the window size. Thus for workloads
like OLTP we see an improvement. For other workloads we see now changes
as no adjustments are made.

The implementation needs some refinement, but the idea appears to work.

Coverage Stats

5321 of 5631 relevant lines covered (94.49%)

0.94 hits per line

ben-manes / caffeine / 1732 / 1
100%
master: 100%

README BADGES
x

Markdown

Textile

RDoc

HTML

Rst

Source Files on job 1732.1 (TERM=dumb)

ben-manes / caffeine / 1732 / 1 100% master: 100%

README BADGES x

Markdown

Textile

RDoc

HTML

Rst

Source Files on job 1732.1 (TERM=dumb)

ben-manes / caffeine / 1732 / 1
100%
master: 100%

README BADGES
x