Performance Engineering Performance Metrics Prof Jerry Breecher Performance

Performance Engineering Performance Metrics Prof. Jerry Breecher Performance Metrics 1

All About Numbers Performance Metrics 2

PERFORMANCE METRICS In this section, a few of the metrics of performance are described. Along with these metrics are some simple equations that connect together the various quantities. We will continue to talk about metrics in a later section. In fact, when we get to talking about queues, we’ll be spending lots of time looking at the interrelations of these concepts. Arrivals ------> SYSTEM ------> Departures T is the length of TIME we observed the system. A is the number of request ARRIVALS observed. C is the number of request DEPARTURES observed. W the ACCUMULATED TIME for all requests within the system - time spent both waiting for and using resources. It is S requests[ Tdeparture – Tarrival] Performance Metrics 3

PERFORMANCE METRICS Arrivals ------> RESOURCE ------> Departures B is the length of time that the resource was observed to be BUSY. Note that the pictures may be equivalent; a system may have only one resource. A request is presented to a system ( in which case it might be called a "job") or is presented to a particular resource. Performance Metrics 4

PERFORMANCE METRICS Metric Description Formula Arrival Rate Y = A / T How many requests arrive per time. Throughput (Departure Rate) How many requests complete per time. X = C / T Utilization U = B / T Fraction of time the resource is in use. Service Requirement Time needed by a resource to handle a request. S = B / C Requests in system Average number of requests in the system. N = W / T Average time a request spends in the system. R = W / C Residence time Performance Metrics Units Requests/second Number in range 0 - 1. Seconds Dimensionless Seconds 5

PERFORMANCE METRICS Example: Requests 3 | | 2 | --------- -------- | 1 | ----------------- | +-------------------------------------- 0 1 2 3 4 5 6 7 8 9 10 Time ( seconds ) What are: Note that requests are serviced simultaneously - it's a multiprocessor? T A C B Time interval = Arrivals = Completions = Time spent servicing = requests Y = A / T X = C / T U = B / T S = B / C Arrival rate Throughput Utilization Service time/request Performance Metrics = = 6

PERFORMANCE METRICS UTILIZATION LAW U = X S Proof: B / T = C / T * B / C Example: The server of dinners in the Cafeteria keeps busy 75% of the time between 12: 00 and 1: 00. During this time 90 people are served a dinner. How long does it take to serve each customer? Performance Metrics 7

PERFORMANCE METRICS Requests 3 | ---- | | 3 | 2 | ------------ | | 2 | 3 | 1 | ----------------------- | 1 | 2 | 3 | +---------------------------------------- 0 1 2 3 4 5 6 7 8 9 10 Time ( seconds ) T Time interval = A Arrivals = C Completions = B Time spent servicing = requests W Accumulated time within system = Y = A / T Arrival rate X = C / T Throughput = = R = W / C Residence Time = U = B / T Utilization = S = B / C Service time/request = N = W / T Average number of = requests in the system. Performance Metrics 8

PERFORMANCE METRICS LITTLE’S LAW N = X R Proof: W / T = C / T * W / C Example: On an average day in the Cafeteria, four people are waiting in the sandwich line. Each of the two sandwich makers can produce a sandwich in one minute. How long must the diners wait in line until they are handed a sandwich? How many sandwiches can be produced in one hour? Performance Metrics 9

PERFORMANCE METRICS Example: One day I was at the Weston Toll Booth on the Mass Pike. There were 10 collectors working; when I joined the line, there were 8 cars ahead of me, and when my turn came it took me 12 seconds to pay my toll. What information can be derived from the above paragraph? What information can NOT be deduced? Example: Recently it was reported that the current U. S. Marriage rate is 6. 8 per 100 adults and the divorce rate is 3. 6 per 1000 adults. It was also reported that 48% of all adults are currently married. Develop a simple model/picture in order to make some sense of these numbers. Performance Metrics 10

PERFORMANCE METRICS Some more practice problems: Example: Packets arrive at a gateway at a rate of 125 packets/second. The gateway takes an average of 2 milliseconds to forward them. What is the Throughput, Service Time, and Utilization. Example: A system has a CPU, Disk A, and Disk B. The throughputs of these devices are XCPU = 35. 48, XA = 15. 70, XB = 19. 6. We also know that the Queue lengths are QCPU = 8. 88, QA = 3. 19, QB = 1. 40. What are the device response times for these devices? Example: On the campus, there a number of graduate students who have varying degree and course requirements. Build a model using duration of time until their degree, as well as the fraction of time in school. What is the service time, residence time, of a graduate student? What is the Performance Metrics 11 throughput of the CS Graduate School?

PERFORMANCE METRICS SUMMARY OF PERFORMANCE METRICS T A C W is the length of TIME we observed the system. is the number of request ARRIVALS observed. is the number of request DEPARTURES observed. waiting for and using resources. is the length of time that the resource was observed to be BUSY. B Arrival Rate Throughput (Departure Rate) Utilization Service Requirement Requests in system Residence time Y = A / T X = C / T U = B / T S = B / C N = W / T R = W / C UTILIZATION LAW U = X S LITTLE’S LAW N= X R Performance Metrics 12