early benchmarks on pre-release Gnocchi v4. includes benchmark comparison between all-ceph v3.x driver versus all-ceph v4 driver. also, shows benchmark using redis+ceph deployment.
2. TL;DR (Executive Summary)
Gnocchi v4 accepts incoming measures faster and is more flexible. It also aggregates
measures quicker with less system requirements.
Not bad.
6. Procedure
1. POST 1000 generic resources spread
across 20 workers, 20 metrics each
a. 20 metrics per POST
b. 1 point/metric per POST
c. 1.2 million points per run, 20K
metrics, 2 granularities x 6
aggregates
2. Wait for backlog to clear
3. Repeat
13. Notes
● ~8.5x improve POST time vs
v3.x
● v4 Redis incoming ~15%
quicker than v4 Ceph incoming
● Significantly more load on
original v3.x test because of
more metricd workers
● v3.x vs v4 POST under similar
load showed closer performance
but v3.x degraded heavily over
time
POST time
14.
15.
16. Notes
● ~12x faster at processing
measures
● v4 tests uses ⅓ the number of
metricd processing agents vs
v3.x test
● v4 redis+ceph deployment
~20% improvement compared
to v4 ceph deployment
● v3.x testing with 18 metricd
and 20OSD failed
Aggregation time