Qingqing Zhou <zhouqq@cs.toronto.edu> writes:
> Yeah, understood. What I can't understand that in this case why it costs
> so much -- without concurrency, the LWLock code path just invloves
> spinlock_lock/unlock and serveral simple instructions?
You executed LWLock 2.6 million times in just under 300ms. If my math is right
that's about 115 nanoseconds per lock or about 300 cycles on a 2.6Ghz
processor.
That sounds like a lot but it's about the right order of magnitude. Was this
on a multiprocessor machine? In which case a big part of that time is probably
spent synchronizing between the processors.
--
greg