I was just trying to suggest a way to reduce the number of SAMxx
switches needed during loading. It looks like the current architecture
doesn't facilitate changing it from processing one row at a time. It
was mentioned that it's probably the number of SAMxx instructions and
not the 64-bit access that causes your performance issues.
Do you have APA or Strobe or some other performance analyzer product
available?
Regards, Gary
On 2015-05-05 14:56, Brite wrote:
These programs don't do I/O. They are called by other programs that do I/O. The
record will be moved from I/O buffer (I remember they are always below 16MB) to
storage above 2GB. Is there performance difference between moving from below
16MB to above 16MB and moving from below 16MB to above 2GB?
