Calibrating I/O

Today a colleague of mine started the short calibration code:

set serveroutput on
declare 
 l_latency number;
 l_iops    number;
 l_mbps    number;
begin
 dbms_resource_manager.calibrate_io(num_physical_disks=>6, max_latency=>12, max_iops=>l_iops, max_mbps=>l_mbps, actual_latency=>l_latency);
 dbms_output.put_line('max_iops='||l_iops||'   max_mbps='||l_mbps||'  actual_latency='||l_latency);
end;
/

... and after some 10 minutes got

Tue Oct 24 13:26:50 2017
Errors in file /opt/oracle/diag/rdbms/xxx/xxx/trace/xxx_cs01_10660.trc  (incident=60493):
ORA-04031: unable to allocate 1049112 bytes of shared memory ("shared pool","unknown object","sga heap(1,1)","ksfdcamem shared I/O read buffer")
Incident details in: /opt/oracle/diag/rdbms/xxx/xxx/incident/incdir_60493/xxx_cs01_10660_i60493.trc
Use ADRCI or Support Workbench to package the incident.
See Note 411.1 at My Oracle Support for error and packaging details.
Errors in file /opt/oracle/diag/rdbms/xxx/xxx/trace/xxx_cs01_10660.trc:
ORA-04031: unable to allocate 1049112 bytes of shared memory ("shared pool","unknown object","sga heap(1,1)","ksfdcamem shared I/O read buffer")
Errors in file /opt/oracle/diag/rdbms/xxx/xxx/trace/xxx_cs01_10660.trc:
ORA-04031: unable to allocate 1049112 bytes of shared memory ("shared pool","unknown object","sga heap(1,1)","ksfdcamem shared I/O read buffer")
Tue Oct 24 13:26:51 2017
Dumping diagnostic data in directory=[cdmp_20171024132651], requested by (instance=1, osid=10660 (CS01)), summary=[incident=60493].
Tue Oct 24 13:26:51 2017
Sweep [inc][60493]: completed
Sweep [inc2][60493]: completed

While there is not much in the subject on Metalink, there exists an excellent blog article. I may add to it the following:

• 1st idea just to flush shared pool did not help - apparently as the system was in use initially and the load taken off just prior to calibration and many allocations already took place
• 2nd idea was to restart and that was enough - when run immediately after start, the process went smoothly.
• after success we 've got

  Tue Oct 24 13:42:51 2017
  Shared IO Pool defaulting to 512MB. Trying to get it from Buffer Cache for process 12832.
  

  Now it seems quite obvious that even with shared pool allocating 2G after a while there is no such amount of free memory in 1M chunks - even in reserved shared pool part (shared_pool_reserved_size defaults to 5% of the shared pool total)