-
Notifications
You must be signed in to change notification settings - Fork 31
Oracle configuration tests
The purpose of this writing is to test and document Rucio performance with various configuration parameters in Rucio configuration, Apache server, Oracle database, and Kubernetes cluster.
Load is generated by three sources:
- Rucio daemons
- Rucio interactive client calls
- PhEDEx to Rucio synchronization scripts
Web UI will possibly be added later.
Metrics observed:
- Response time to interactive client commands
- Number of simultaneously open user sessions (it is limited in Oracle for performance reasons)
- Errors rate
client: request_retries
server:
Configuration scenarios: On Rucio side:
-
Initial configuration with DRCP disabled
-
Initial configuration with DRCP enabled
-
Reduce pool_timeout 600 -> 30
-
Reduce number of servers to 1 rucio, and 1 auth . On Oracle side:
-
Franck changed pooled_max_think_time to 1200 https://docs.oracle.com/database/121/ARPLS/d_connection_pool.htm#ARPLS65622 Run 30 client-perf tests. 5 . Re re-create cluster with three servers , same settings. Run 30 client-perf tests.
-
Same cluster and settings, run sync (two sync jobs in the beginning). Monitor sessions.
-
Add client performance test
- https://github.com/rucio/rucio/blob/master/lib/rucio/client/baseclient.py - Rucio code handling authentication
- https://www.oracle.com/database/technologies/appdev/oracle-net-services.html - Oracle performance resources