2 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid.dtd">
4 <AS id="AS0" routing="Full">
6 <!-- Use the pstate mechanism to encode the boot/shutdown time and energy -->
8 <!-- That may be seen as a misuse of the pstate mechanism and crude hack,
9 but this is efficient: we can do what we want without too much burden,
10 and since the whole intelligence is at user level, we do control the model.
12 The same could be done to encode the time and energy to switch
13 between pstates. These times seem negligeable in our
14 experiments, but if they are important to you, now you know how
19 0: p1 100 MFlops/s, [idle: 95W -> full burning: 200W]
20 1: p2 50 MFlops/s, [idle: 93W -> full burning: 170W]
21 2: p3 20 MFlops/s, [idel: 90W -> full burning: 150W]
22 * 3: booting up was measured to take 150s and 18000J.
23 So we create a pstate with 1 flop/150 s so that the
24 application burns 1 flop in that virtual pstate to
26 Multiply everything by 1000 or whatever if this trick is
27 too painful to you, that's the same by the end.
28 The energy consumption is 18000/150=120W
29 * 4: shuting down was measured to take 7 s and 770 J
31 Please note that we tried to put sensible values in this file
32 but you should still take them with a grain of salt. If you
33 want a realistic simulation, you have to actually benchmark
34 your platform and application, and use the values that are
38 <host id="MyHost1" power="100.0Mf,50.0Mf,20.0Mf, 0.006666667f,0.1429f" pstate="0" >
39 <prop id="watt_per_state"value="95.0:200.0,93.0:170.0,90.0:150.0, 120:120,110:110" />
40 <prop id="watt_off" value="10" />
42 <host id="MyHost2" power="100.0Mf" >
43 <prop id="watt_per_state" value="100.0:200.0" />
44 <prop id="watt_off" value="10" />
47 <link id="link1" bandwidth="100kBps" latency="0"/>
48 <route src="MyHost1" dst="MyHost2"><link_ctn id="link1"/></route>