Você está na página 1de 2

CS490WW/CS5590WW/ECE5590WW HW4 due Tuesday July 26

1. The designer of a Web site wants to build the site using 10 servers. The ratio between MTTR and MTTF is = 0.1. The support sta is capable of servicing 2 servers concurrently if they fail. Each server has Dcpu = 0.1s and DIO = 0.3s, and can handle a maximum of 10 requests. The overall arrival rate to the site is 20 requests/s. (a) What is the average response time Rup , the fraction of lost requests Fup , and the throughput Xup for the Web site? 2. Consider the following number of HTTP requests per second measured at two Web sites over a period of 10 months Table 1: Number of requests during peak second at server 1 and 2 Requests Per Second Peak (Y 1t ) Requests Per Second Peak (Y 2t ) 517 425 551 411 580 438 625 466 675 450 777 463 984 507 1,275 513 2,106 532 4,315 515

Month (t) 1 2 3 4 5 6 7 8 9 10

(a) Use linear regression to nd a function f that predicts the number of HTTP requests for each server. (Note: you must decide if a linear or logarithmic expression best ts the data) (b) Calculate the coecient of determination for each function you have derived. (c) With the functions you have derived, generate the forecasting trend for the next 4 months. 3. Consider data collect from an HTTP log shows that a server had received 24,000 requests during 1 hour. Using a clustering algorithm, the requests can be partitioned into 3 classes: le sizes 22 KB which use a total CPU time of 1,500s based on 12,000 requests logged le sizes from 22 KB to 1 MB which use a total CPU time of 950s based on 8,400 requests logged le sizes greater that 1MB which use a total CPU time of 420s based on 3,600 requests logged The CPU utilization during the measurement period was determined to be 84% (a) What is the service demand at the CPU for each class of request? (b) What is the average number of HTTP requests simultaneously being serviced at the CPU for each class? 4. A Web site wishes to achieve reliability of 99.999% using a cluster of cheap and unreliable servers. If each server has a reliability of 88%, how many servers should be used in the cluster? 1

5. Consider requests are logged at a server at instants 1, 2, 3.4, 3.8, 6, 6.2, 6.5, 6.8, 6.9, 7.0 10, 13, 13.2, 13.6, 13.7, 13.9, 16, 25, 30, 34, 34.2, 34.6, 35, 35.3 during an interval equal to 36 seconds. (a) Calculate the burstiness parameter b for n = 35 6. Consider software that is aging at a rate of ag = 1/250. The rate of failure from the robust state is f 1 = 1/700, and the rate of failure from the degraded mode is f 2 = 1/25. The recovery rate is rec = 1 and the rejuvenation rate is rej = 1/10. Assume that the software aging is load independent and the server must be in robust mode to be considered operational. (a) All servers have innite population arrivals with xed service rate at 0.8 requests/s and innite queue. Determine the number of servers needed so the site functions in robust mode 99.999% of the time.

Você também pode gostar