week9a - androbenchcsl.skku.edu/uploads/ece5658s17/week9a.pdf · 2017. 5. 1. · workload sc d...
TRANSCRIPT
![Page 1: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/1.jpg)
![Page 2: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/2.jpg)
![Page 3: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/3.jpg)
![Page 4: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/4.jpg)
![Page 5: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/5.jpg)
![Page 6: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/6.jpg)
![Page 7: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/7.jpg)
![Page 8: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/8.jpg)
![Page 9: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/9.jpg)
![Page 10: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/10.jpg)
![Page 11: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/11.jpg)
![Page 12: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/12.jpg)
System cycles as a percentage of total cycles
![Page 13: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/13.jpg)
• Memory penalty for system instructions• Mach > Ultrix
• The number of instructions executed in virtual memory system• Mach > Ultrix
• Relative instruction cost• Mach < Ultrix
• Overhead of Mach’s IPC is responsible for a small portion of overall system overhead
![Page 14: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/14.jpg)
![Page 15: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/15.jpg)
• System MCPI: Ultrix < Mach
• User MCPI: Ultrix ≈ Mach
system instruction locality
![Page 16: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/16.jpg)
• It can be measured by cache and TLB miss rate
competition between the user and system
![Page 17: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/17.jpg)
• High MCPI -> cache is full• Difficult to be improved by the increment of
associativity
• The associativity is more helpful to Ultrix than Mach• Mach’s MCPI is higher than Ultrix
system self-interference
![Page 18: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/18.jpg)
• Block operations incur a larger overhead for programs running on Mach than on Ultrix
block operation
Block operation type In kernel In UNIX Server
Mach Part of VM and IPC system Part of file system
Ultrix VM and file system None
![Page 19: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/19.jpg)
• System write buffer stalls per instruction are generally higher for Mach than for Ultrix
streaming writes
![Page 20: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/20.jpg)
page mapping strategy
![Page 21: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/21.jpg)
![Page 22: week9a - AndroBenchcsl.skku.edu/uploads/ECE5658S17/week9a.pdf · 2017. 5. 1. · workload SC d egrep yaCC gec c o press espresso lisp eqn tott fpppp doduc liv t 0m cat v Instruction](https://reader035.vdocuments.mx/reader035/viewer/2022071513/6134bddddfd10f4dd73beccb/html5/thumbnails/22.jpg)