|
|
|
We observed a significant processing burden in a performance test in calls to the 64-bit arithmetic routines in the function that manages the data file I/O through the data cache. For members of superfiles, it is also possible for index header writes and space management node writes for the “index” I/O to ultimately get to disk through this same data cache processing code path.
This code was modified to skip the 64-bit arithmetic routines for non-huge files, and simply perform 32-bit arithmetic on the low order file addresses. We also observed a file address computation that could be removed completely for both huge and non-huge files.
