Thanks for your reply. 1)I really need to know the link bandwidth, so I can calculate how much time to transmit certain amount of data. If it's a gigabit link, why are those rxBandwidths and totBandwidths just tens of bits per second? I looked through the source code etherlink.cc, does the link bandwidth equal sim_freq/ticksPerByte? 2)Is the execution time on testsys equal to (drivesys.cpu.numCycles/2GHZ)? In M5 simulator, does 2G mean 2*2^30 or 2*10^9?
Xin ---- Original message ---- >Date: Wed, 28 Nov 2007 06:50:53 -0800 >From: "Steve Reinhardt" <[EMAIL PROTECTED]> >Subject: Re: [m5-users] statistics on M5 2.0b3 >To: "M5 users mailing list" <m5-users@m5sim.org> > >On Nov 27, 2007 11:52 PM, <[EMAIL PROTECTED]> wrote: >> Hello, >> >> I have run a client-server program, use the sample fs.py >> (2GHZ atomic cpu with 512MB memory)and have the following >> questions about the simulation statistics. >> >> 1) eth0: 1000F mbps >> what does the 'F' stand for? full duplex? Can I think it >> means the bandwidth between client and server is 1000mbps? >> However, in m5stats.txt, >> drivesys.tsunami.ethernet.rxBandwidth=19(bits/s) >> drivesys.tsunami.ethernet.totBandwidth=34(bits/s) >> testsys.tsunami.ethernet.rxBandwidth=15(bits/s) >> testsys.tsunami.ethernet.totBandwidth=34(bits/s) >> Does anybody explain the above numbers? > >It's a gigabit link, you're just not using it much. > >> >> 2) drivesys.cpu.not_idle_fraction=0.000894 >> drivesys.cpu.numCycles=544382753 >> drivesys.cpu.num_insts=544372834 >> >> host_inst_rate=861547(inst/s) >> host_seconds=1266.56 >> host_tick_rate=240466519018(ticks/s) >> sim_freq=1000000000000(ticks/s) >> sim_insts=1091201381 >> sim_seconds=304.565430 >> sim_ticks=304565430390500 >> >> testsys.cpu0.not_idle_fraction=0.000898 >> testsys.cpu0.numCycles=546839739 >> testsys.cpu0.num_insts=546828547 >> Why are both not_idle_fractions so low? about 0.089%? >> What is the difference between host_tick_rate and sim_freq? >> What is the difference between host_seconds and sim_second? >> How to compute the time(seconds) used to execute >> instructions on each system? I have tried host_inst_rate >> to be divided by testsys.cpu0.num_insts. However, the >> result is more than double sim_seconds. > >All of the "host_*" stats refer to the machine the simulator is >running on; e.g., host_seconds is the wall-clock run time of the >simulator. Those stats have nothing to do with the simulation itself. > >sim_seconds is the number of seconds simulated, as the comments in >m5stats.txt indicate. > >Steve > >> >> Xin >> _______________________________________________ >> m5-users mailing list >> m5-users@m5sim.org >> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users >> >_______________________________________________ >m5-users mailing list >m5-users@m5sim.org >http://m5sim.org/cgi-bin/mailman/listinfo/m5-users _______________________________________________ m5-users mailing list m5-users@m5sim.org http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
