G
George Potemkin
Guest
> But in one database, the values of Latch timeouts are almost always in the red zone The cost of one timeout is unknown. Is it the -nap 10 or the -napmax 250 (or whatever is used by your customer)? The difference is huge. I'd change the -napmax to the -nap just to check how it will affect the latch timeouts. Significant increasing of timeouts would mean that a bottleneck on the latches is real. Also I guess the -nap 1 would be a right thing to use. > 128 CPUs is a very, very bad thing for a database server. What is a percent of time when one Progress session uses the latches? 5%. Maybe 10% at most. So the most processes that are using CPUs right now are not locking the latches. The rest (~ 6 processes = 5% of 128 ) served by CPUs at the moment: only one process can lock a regular latch at time, a few processes can /try/ to lock the same latch. But if they failed then will sleep (nap 10 milliseconds) the thousands times longer than latch lock (~50-100 nanoseconds) leaving CPU time for the processes that don't need the latch locks. And if a box is running many databases then the large number of CPUs will not be a bad thing at all.
Continue reading...
Continue reading...