Never mind. I figured it out. MAX_IOPS and MAX_MBPS are only meant for
throttling the tenants. They have nothing to do with calibration. That
leads us to having one value that optimizer uses to determine automatic
degree of parallelism and something completely different on the tenant
level, restricted by the tenant parameters. There is also the 3rd value,
if we count the throughput parameter from AUX_STATS$ table. Oh, well...
On 5/17/19 7:20 AM, Mladen Gogala wrote:
Oracle 19c has parameters MAX_IOPS and MAX_MBPS which set the values that used to be set by using dbms_resource_manager.calibrate_io procedure. Does that mean that we can now use SLOB or bonnie++ to determine the IO system performance instead of running the notoriously unreliable calibrate_io? From the documentation, I got the impression that this is sort of a hard limit:--
https://docs.oracle.com/en/database/oracle/oracle-database/19/refrn/MAX_IOPS.html#GUID-7D89F457-0CE6-4152-98EA-7F1BE4E0ADC7
On the other hand, calibrate_io was previously used only for determining the automatic degree of parallelism and wasn't set as a hard limit. The new mechanism will make the processes wait should the IO rate exceeds the one defined by parameters. Will the optimizer use this or SYS.RESOURCE_IO_CALIBRATE$ to determine the automatic degree of parallelism, where that is enabled? We can have one value in the parameters and something completely different in SYS.RESOURCE_IO_CALIBRATE$.
