Approximation of Worst-case Execution Time for Preemptive ...
Approximation of Worst-case Execution Time for Preemptive ... Approximation of Worst-case Execution Time for Preemptive ...
Statistics Gathering • Problem: choose representative traces • Solution: – profile different input sets – conservative approximation • The tests confirmed a certain homogeneity within different execution traces for the same tasks Matteo Corti, ETH Zurich, LCTES 2000 14
Cycles Per Instruction (CPI) … • The instruction length can be divided in several components: – ICP: infinite cache performance (CPU busy and stall time) – FCE: finite cache performance (effects of memory hierarchy) CPI = ICP + FCE CPI = busy + stall + FCE execunit + stallunit CPI = + stall pipeline + parallelism CPI = ... FCE Matteo Corti, ETH Zurich, LCTES 2000 15
- Page 1 and 2: Approximation of Worst-case Executi
- Page 3 and 4: Environment: User Needs • Complex
- Page 5 and 6: Problem Description • Admission t
- Page 7 and 8: Other approaches • Longest path:
- Page 9 and 10: Longest Path ... instr op op op ins
- Page 11 and 12: Instruction Length • Preemption,
- Page 13: Performance Monitor • Not specifi
- Page 17 and 18: Testing the Predictor • First pha
- Page 19 and 20: Results: Approximations • Worst c
- Page 21 and 22: Results: Real Applications • Only
- Page 23 and 24: Comments … • Performance monito
- Page 25: Conclusions • The WCET can be app
Cycles Per Instruction (CPI) …<br />
• The instruction length can be divided in<br />
several components:<br />
– ICP: infinite cache per<strong>for</strong>mance (CPU busy and stall time)<br />
– FCE: finite cache per<strong>for</strong>mance (effects <strong>of</strong> memory<br />
hierarchy)<br />
CPI = ICP +<br />
FCE<br />
CPI = busy + stall + FCE<br />
execunit + stallunit<br />
CPI =<br />
+ stall<br />
pipeline<br />
+<br />
parallelism<br />
CPI = ...<br />
FCE<br />
Matteo Corti, ETH Zurich, LCTES 2000 15