>>>>> "Matthew" == Matthew Costello <matthew@stripped> writes:
Matthew> I've noticed a slowdown as well when upgrading to 3.22.26a,
Matthew> although in this case the upgrade was from 3.21.33c. This is
Matthew> using Red Hat Linux . Vmstat(8) shows the number of context
Matthew> switches under moderate load to be 200,000 per second with
Matthew> the new RPM, which is an insane number. I expect this is
Matthew> caused by some sort of lock contention. Compiling from source
Matthew> lowered the context switches to 70,000 per second, which is
Matthew> still way too high.
Matthew> /matthew costello
>> >>>>> "Viren" == Viren Jain <viren@stripped> writes:
Viren> After upgrading my previous MySQL installation, which was ver 3.22.22 and
Viren> had a 2 meg mysqld binary (which to me seems important because on my new
Viren> installation, the binary is only 1 meg or so), to ver 3.22.26a, there seems
Viren> to be a fairly significant performance hit. I'm not quite sure this is due
Viren> to any changes between those versions, but might be that the two mysql's
Viren> were compiled with different options. This is the configure line used on
Viren> the 3.22.26a installation (the configure line for the 3.22.22 is unknown):
Any change you can run the MySQL benchmarks with 3.21 and 3.22 ?
3.22 should be faster in most cases...
I can't also not understand why a recompile should affect context
switches in any way...