There was a performance issue due to the high activities and some inefficient coding. The system(Solaris) load was high. To mediate the issue, we dropped the spin on the latch. Right after that, the non-responsive screen started to move, an unix guy saw that happen. So the DBAs are asked to keep the spin down all the time, which we did. But we saw that the lower spin caused very high nap number and other unbalances within the DB. Does high spin number or low spin number really matter that muhc?
thanks.
You should provide your version number and the actual before and after spin values in order to get meaningful response.
this is a Solaris box
8 socket, 16 cores, 32 threads
the spin was 50000, we lowered it to 10000.
Progress 10.1B with latest patch and hotfix.
The default values for -spin are almost always way too high.
There were very significant changes made to the latches in 10.1C
I currently advise people who are on more or less current releases to use a relatively low value such as 5,000. 10,000 isn't bad either.
There is (very) bogus advice floating around in the kbase and in some white papers that advises using a -spin of N x #CPUs where N is a value such as 10,000 or 20,000. This is bad advice. Do not follow it. It came from something that Gus said way back when and which he almost immediately retracted. Yet it lives on. (It you don't believe me feel free to ask Gus, he will back me up.)
IMHO the minimum value for -nap is also way too high. I think it made sense 15 years ago when CPUs were 10x slower but, in todays world it should be measured in microseconds not milliseconds. The best you can do now is to set -nap to 1 (only on the command line, promon won't let it go below 10). I'd like to be able to set it to 0.01. Or lower.
thanks, Tom for the helpful information. I like the keeping nap short option.
would you keep the spin low though when the timeout increases 10 fold? or you think this is can be addressed by shorten the nap time? essentially, they are the same thing.
-nap & -napmax are different ends of the spectrum.
-nap is the initial nap, the nap time increases by doubling as you "spin
out" until it gets to -napmax.
I haven't had a whole lot of luck modifying either -- sometimes I get a
bit of improvement but not a lot. When I have had success it has been
with settings like -nap 1 and -napmax 500. I think I'd have better
luck if -nap could be smaller but, since I can't try it, I don't really
know
Your mileage may vary.