Micheal Mukherji wrote:
On Mon, 13 Dec 2004 08:43:28 -0600, Ron Johnson <ron.l.johnson@cox.net> wrote:Remember, though, that a ns is 1 billionth of a second, and so a 3GHz CPU will onll does 3 cycles in a ns. Unless you are coding to the bare metal with a *minimal* OS, like DOS, you can't do anything useful in 3 clock ticks.On a 1.4 GHz processor, I use JVMPI( an abbreviation for Java Virtual Machine Programming Interface). It has a method called GetThreadCPUTime(), which it claims gives thread execution time in nanoseconds. He does not mention any hardware support. How do you think he is going to provide it?? I am getting values in tens of nano seconds also..
This hilights the difference between 'accuracy' and 'precision'. When an interface claims to offer nanosecond time information, it's not making any claim about how accurate its information will be. So, for example, you might be able to read values like the following from the timer, if you have a sufficiently fast program: 21223232000 21223233000 21223234000 21223235000 21223236000 Those values are very precise (down to the ns), but in general they're not any more *accurate* than a *microsecond* timer could be. I suspect that people write APIs in terms of nanoseconds so that, if/when the hardware/os actually supports measuring time that accurately, the API doesn't need to change. -- Christian Convey Computer Scientist, Naval Undersea Warfare Center Newport, RI