SFWRENG 4E03 Lecture Notes - Lecture 9: Overclocking, Markov Chain, Poisson Distribution

• School: McMaster University
• Department: Software Engineering
• Course: SFWRENG 4E03
• Professor: Kemal Ahmed

Document Summary

M/m/c provisioning 3 e. g. 1) 4. Example continued: split arrivals into 2 streams (cid:315)75 (cid:315)75 (100) (100, overclocked server (cid:315)150 (200, only overclock when more than 1 task in server (cid:315)150 (100/200) Each job accumulates cost at /min , i. e. the rate at which it costs. The expected cost as a result for the entire system will be: /min e[n], e[n] ~ avg jobs in sys. : arrival rate, i. e. , e[r](cid:540) (cid:314)little"s law (cid:314) , e[n] Both streams for a are m/m/1 queues, i. e. e n. E[nc] = m/m/2, (cid:540) = 150, (cid:541) = 100. Doubles its capacity when it experiences extra load. Review slides under ctmc if you missed classes. Given (cid:540) and (cid:541), what should c be so pq < . E[rq] = e[rq | waits in queue] pq + e[rq | doesn"t wait in queue] (1 pq) If q doesn"t enter queue, request time is 0, i. e. e[rq] = 0, so simplify the equation.