6. Consider
● 2 5+ IOPs per update (read-modify-write)!
● 1 RRD per data source (storeByGroup=false)
● 100,000s of data sources, 1,000s IOPS
● 1,000,000s of data sources, 10,000s IOPS
● 15,000 RPM SAS drive, ~175-200 IOPS
7. Also
● Not everything is a graph
● Inflexible
● Incremental backups impractical
● ...
8.
9. Bottleneck!?
● We need to be collecting even more!
● We need to be collecting more frequently!
● The Internet of Things is upon us!!
10.
11. But can’t we … ?
● Serialize RRD writes?
● Cache?
● Distribute the RRDs?
● … ?
12.
13.
14. How about:
● Distributed, decoupled architecture
● High throughput
● Horizontally scalable
● Pluggable, extensible graphing
● Facilitate new forms of analytics
● More?
Starting Over Can Be Fun!
15. Observation #1
We collect and write a great deal; We read
(graph) relatively little.
We are read-optimized.
30. Gist
● Samples stored as-is.
● Samples can be retrieved as-is.
● Measurements are aggregations calculated
from samples (at time of query).
31. Samples vs. Measurements
sam·ple
/ˈsampəl/
noun
1. a small part or quantity intended to show what the whole is like.
"investigations involved analyzing samples of handwriting"
synonyms: representative, illustrative, selected, specimen, test, trial, typical
meas·ure·ment
/ˈmeZHərmənt/
noun
1. the action of measuring something.
"accurate measurement is essential"
synonyms: quantification, computation, calculation, mensuration