[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [K12OSN] 64 bit ltsp vs. virtualization



On Tue, 24 Jun 2008, James P. Kinney III wrote:

>
> On Tue, 2008-06-24 at 07:05 -0400, Julius Szelagiewicz wrote:
>
> > 	I have one application - DB/C language interpreter, that all the
> > users run, that shows eightfold increase in speed between 32bit and 64bit
> > versions.
> That is very odd. 64-bit does not provide speed. Speed is handled by the
> clock rating of the core. The bit count determines the size of the
> addressing area that can be directly accessed for RAM. Internally, the
> cpu's don't have enough cache to be affected by the bit size.
> An 8-fold speed increase is tremendous. If this is being run on a
> multi-core machine, the speed increase may be more related to better
> threading control in the 64-bit system than the 32-bit (assuming the
> same machine). The 64/32 bit system like Opterons and Xeons are more
> efficient in 64-bit mode as it's a single clock tick to fill buffers but
> it takes an extra tick every 2 fills for 32-bit mode (has to store 2
> addresses per fill in 32 bit vs. 1 per 64 - cache writes take a tick).

James, it's not that odd - the execution speed is affected by the size of
a memory fetch - double per cycle compared to 32bit, machine level moves
and compares are also doubled in speed. I was expecting a three- to
five-fold increase in speed, because all the programs perform a lot of
I/O, but even that was affected - the prefech reads into the I/O cache got
faster.
julius


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]