[理工] [計組] 97台大資工

[boy5548 (小YO)]

97台大資工計組第2題(d)題目重點如下：

An application has been fully parallelized to run on a SMP system, but for

an 8-processor run, 50% of time a processor has to stall because the processor

is waiting to access the disk.There is only one disk in the system.Assume that

the OS issues one disk accesses at a time.

(d) Suppose that the application runs for 100 sec when only one processor is

used,how many sec would tha application requires to run on the 8-processor

SMP system? (Note that the 100 sec is the wall clock time,measured from the

beginning to the end of the application.)

--

張凡解答這樣寫：Suppose x is the I/O time, (100-x)/(0.5*8)=x => x=20

execution time = 20+20 = 40 sec

其實看不太懂這個式子...想請教一下各位，謝謝!!

--
※ 發信站: 批踢踢實業坊(ptt.cc)
◆ From: 118.171.44.204
→ privatewind:其實我覺得這題張凡 解錯了...
→ boy5548:那請問一下喔 你覺得這題要怎麼解?
→ boy5548:他和高銘解法不同...
→ privatewind:d 我算22.22 s, x+x= 8cpu-excution time
→ privatewind:x+8x= 1cpu-time =100, x=11.11
→ privatewind:11.11*2=22.22
→ zelkova:可是p大.. 有50%要stall, 所以x+8x是不是要改成x+4x?
→ zelkova:只是這樣答案好像跟張凡一樣@@
→ privatewind:我的x是IO time
→ privatewind:50%要stall 這個50% 就是x佔去的時間
→ privatewind:他的stall time 應該就只有指 IO time 如果還要考慮
→ privatewind:memory stall 那不能算....因為他給的時間是wall time
→ zelkova:有道理 你說服我了@@
→ zelkova:我也不清楚張凡算式的來龍去脈..
→ privatewind:把張凡 式子去掉0.5 就是我的算法，他認為cpu time
→ privatewind:兩者是分開的...
→ privatewind:但實際上IO time 應該包含在CPU time 中
→ privatewind:                         ^^^^ CPU wall time
→ boy5548:p大..其實我不太懂你的算式..可以解釋一下嗎?? 謝謝
※ 編輯: boy5548         來自: 114.27.125.101       (01/31 20:19)
推 nypgand1:#1D81_CQa 我之前問過 也是沒結果
→ privatewind:cpu execution time = CPU Wall time
→ privatewind:簡單講就是 我的stall time = io time
→ privatewind:他也說了 50%的stall 是因為IO
推 babygoat:p大講的應該是execution time吧  張凡之前的章節有講過
→ babygoat:execution time= CPU_time + I/O time所以他的算法才會分
推 lsy77613:請問p大,為什麼(100-x)/(0.5*8)這個式子會等於x
→ lsy77613:不是應該等於新硬體的執行時間嗎?
→ sneak: 可是p大.. 有50% https://daxiv.com