D
dbeavon
Guest
>> So how fast is fast enough? Maybe 1/10 or 1/5 what it is now. Or at least 1/#CPU available. I guess I don't like any CPU bottleneck that seems somewhat artificial (where the operation takes 50 seconds because it is only aware of one of my cores out of 8). It doesn't seem like there is any technical reason why the schema tables should be created one at a time rather than concurrently. Ideally some of the work could be done concurrently, off to one side, and then synchronized at the very end and merged into the final results. Even if we allow that the work must be done entirely on one core, then I still find it fairly odd that this takes 50 seconds on a 3.2 GHz CPU. I can't imagine what type of validation work it must be doing to take that long! I'm guessing much of the work is unnecessary and could be totally by-passed in the specific scenario that I'm describing (ie. because the database is new, is empty, and is not even served). I'm guessing that much of the work might involve checking for database users, schema locks, pre-existing schema, and pre-existing data. We have a fairly large project with ~25,000 compiles (.p and .cls) and ~50,000 source files. And the compiling of the source takes only about 30 seconds on eight cores. So it is unfortunate when the initial creation of an *empty* database takes almost *twice* that time.
Continue reading...
Continue reading...