On Thu, Aug 23, 2012 at 7:36 PM, Alireza Haghdoost <[email protected]>wrote:
> Hi Jim > > It is very nice experiment ... Did you run that to get OS overhead ? > What was the difference between cpu_get_ticks() and rdtsc ? > > cpu_get_ticks() is QEMU function (modified to support simulation clock) while 'rdtsc' is an x86 instruction executed in VM to read CPU ticks. - Avadh > > Thanks > Alireza > > > > > > On Tue, Apr 17, 2012 at 2:58 PM, <jims at cs.umd.edu> wrote: > > Hello, > > > > I'm interesting in timing OS overhead for disk accesses. As I've > discussed > > with Avadh in private emails, I've extended the MARSS_DELAY_IO code in > > qemu/hw/ide/core.c to support an SSD simulation. At this point, I want to > > understand how much OS overhead is introduced for the operations to my > > disk simulation. > > > > I'd like to do the following: > > > > * Have a C++ program that does a small file read (and make sure the file > > is not cached in RAM ahead of time). > > * Do rdtsc before and after the read (times A and D). > > * On the hardware side, whenever I receive the disk IO command to my mods > > in ptlsim.cpp, I'd like to grab the time stamp counter at the beginning > > and end of the access (times B and C). > > * To get the OS overhead, compute B-A and D-C. > > > > > > My questions are: > > > > 1. What variable in PTLSim or QEMU should I grab that is equivalent to > the > > rdtsc instruction? > > > > 2. Just to confirm, does the rdtsc instruction work as expected in > > simulation mode (i.e. it gets updated every simulated clock cycle). > > > > Thanks, > > > > Jim Stevens > > > > _______________________________________________ > http://www.marss86.org > Marss86-Devel mailing list > [email protected] > https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel >
_______________________________________________ http://www.marss86.org Marss86-Devel mailing list [email protected] https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
