-
Type: Task
-
Status: Resolved
-
Priority: Major
-
Resolution: Fixed
-
Affects Version/s: None
-
Fix Version/s: None
-
Component/s: Nuxeo Bench
-
Tags:
-
Sprint:nxcore 10.3.3
-
Story Points:5
The read REST requests are slower after LTS 2016.
It sounds that the minimum response time increase from 3ms to 10ms.
Looking at the continuous benchmark for 2 MongoDB in Read simulation the response show increases:
- 18w11 Today we are at 17.5ms (+/- 14) mostly around 16ms
- 17w37 From 12.4ms (/
11) to 15.6ms (/ 21). - 17w02 was at 7ms (+/-4)
- 16w49 6.9ms (+/-8) LTS 2016
Also, we can see in LTS 2016 that with 4 nodes we can have even had 2.7ms (/- 1) vs LTS 2017 at best 11ms (/- 14).
A very low response time is needed to keep very high sync throughput.
We need to understand exactly where these incompressible 10ms are spent.
Possible cause:
- tomcat maxThreads changed from 150 to 20 ->
NXP-23318 - audit entries is saved into chronicle log first
- mongodb directory slower than sql
- ...
- is related to
-
NXP-23016 Fix infinite cross-instance cache invalidations in cluster mode
- Resolved
-
NXP-25065 DirectoryCache does too many invalidations
- Resolved
-
NXBT-1731 Investigate on REST performance regression on FT 9.1
- Open
-
NXBT-2054 Run a milestone benchmark for 9.10 and analyze results
- Resolved