You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
@manolama , the issue is current implementation of MongoDB adapter is it is using 100 connections by default. Even though "100 connections is enough", it bites when one tries to benchmark with lots of clients (e.g. 500 or 1000) on a multicore machine (e.g. 70 cores).
On contrary, JdbcDBClient uses connection-per-client approach. The implementation difference makes benchmark result comparison obscure.
What do you think of creating a common property db.pool.size that would designate the number of connections in a connection pool?
Certain implementations (e.g. Mongo) would just leverage existing implementation (e.g. maxPoolSize url property for Mongo). The others might need a connection pooler (e.g. HikariCP for JDBC).
Having a common property would enable comparison in a controlled environments and it would simplify results comparison.
PS. I think it makes sense to use db.pool.size equal to the number of YCSB threads by default.
'mongodb' client is not honoring the maxPoolSize=XX for the number of client connections. Instead it is opening a connection per YCSB client thread.
The text was updated successfully, but these errors were encountered: