Difference between revisions of "Ceilometer/AlarmEvaluatorJuno"
(→Statistics over sliding windows) |
(→Statistics over sliding windows) |
||
| Line 3: | Line 3: | ||
===Map-reduce approach=== | ===Map-reduce approach=== | ||
| − | [[File:EvaluatorMapReduce.png | + | [[File:EvaluatorMapReduce.png]] |
| + | <br/><br/><br/><br/> | ||
| + | ===Increment/decrement approach=== | ||
| + | [[File:EvaluatorIncDecrement.png]] | ||
| + | <br/> | ||
| − | |||
| − | |||
| − | |||
| − | |||
Statistics where increments/decrements are possible: | Statistics where increments/decrements are possible: | ||
* count, | * count, | ||
Latest revision as of 19:11, 16 May 2014
Statistics over sliding windows
Two approaches to improve the alarm evaluator were discussed at the summit.
Map-reduce approach
Increment/decrement approach
Statistics where increments/decrements are possible:
- count,
- sum,
- average,
- stddev
Statistics with special implementation:
- min/max : needs O(log n) space (probably in DB), where n = window size
- percentile : classical statistics query
- cardinality : approximate algorithm, extremely efficient
Map-reduce approach covers all these statistics.
