R
Rob Fitzpatrick
Guest
A proserve isn't necessarily the answer. I tried various combinations and permutations of Type I, Type II, online, offline, different block sizes, etc. Some of them made less difference than I expected, maybe because the process is CPU-bound and there's a limit to what my single-hamster-driven machine will do. You're using 4 KB block size. You didn't mention the contents of SimpleStructure.st. Are all the objects put into the schema area? Or are they put in application areas? Type I or II? *In theory*, using Type II will do more I/O as each object occupies a minimum of one cluster (8 or 64 or 512 blocks), even if empty. As an example, when I load my schema into the schema area of an empty 4 KB database, the area grows from 2 MB to 10 MB. But when I have the tables and indexes assigned to Type II areas with 8 BPC, I get a 6 MB schema area, and areas 7 and 8 are 17 and 30 MB respectively. There is more I/O to do. That alone was good for about a 6.5 second difference (Type I being faster). So try loading a .df with no area assignments; if your schema has them already, grep them out to create a new .df and load that. Add -r on your command line. Move your truncate bi before the load; try specifying -biblocksize 16 -bi 16384. See what difference those changes make for you.
Continue reading...
Continue reading...