Module 4 - Capacity Planning & Queuing Analysis Flashcards
Which kind of forecasting methods are based on management intuition, sales force observations, a customer focus group, etc?
Note: They tend to be affected by biases, desires, and recent events and therefore are only useful for very innovative products or services
Judgmental forecasting
Which kind of forecasting methods are based on when suppliers and producers work together to predict demand, including taking actions to coordinate product launches and promotions?
Collaborative forecasting
A forecasting model is chosen based on its ability to accurately predict what?
the past
The four most important characteristics of forecasts are:
- The forecast will be wrong. (ie - it will rarely be exactly equal to actual demand)
- A forecast is more accurate for larger groups
- Forecast accuracy is inversely proportional to the time horizon (ie - more accurate in the short term than in the long term)
- A forecast is useful only when a realistic range of uncertainty is provided
What is the range of uncertainty?
The span of values within which the forecaster expects actual demand to fall (with a high degree of confidence)
What is the the systematic component of the forecast?
What greek letter?
How is it calculated?
the forecast
denoted by the Greek letter mu, μ
It is, in effect, the mean value of the predicted demand
What is the the random component of the forecast?
What greek letter?
What is it based on?
How is it calculated?
the margin of uncertainty.
denoted by the Greek letter sigma, σ
It is based on how well the forecasting model would have performed under conditions similar to those anticipated for future demand.
It is quantified as the standard deviation of the predicted demand
Formula for coefficient of variation (CV)?
Why is it useful?
CV = (standard deviation/mean)*100
= (random component / systemic component) *100
It tells us how much variation there is relative to the predicted demand
The intention of a forecaster is to provide two components associated with a future demand prediction. What are the two components?
systematic component and random component
When demand cannot be predicted without significant uncertainty, responsiveness needs to be:
(a) fast
(b) doesn’t matter
?
a. fast
it should be a key element of the production system when there’s less certainty
At a call center where calls for troubleshooting are categorized by reason, why would the CV for total calls per day tend to be less than the CV for calls for each reason?
Because the total calls would combine the uncertainty across each reason.
What are the four terms are critical in a capacity analysis?
(1) Effective Capacity
(2) Unit Load
(3) Utilization
(4) a bottleneck
Effective Capacity is:
the amount of work (flow rate or throughput) that a resource produces
Unit Load is:
the time to process one flow unit
(the inverse of capacity)
Utilization is:
the percentage of time a resource is working (calculated as the time spent processing divided by the capacity)
A bottleneck is:
the slowest resource in a process with multiple resources – the process cannot output units faster than the bottleneck
it’s the resource with the highest utilization
The size of a capacity buffer is based on:
1- the amount of demand uncertainty
2- the service time variation
3- the targeted performance of the resource
Utilization formula
How much is a worker being utilized?
the ratio of demand to capacity
= demand / capacity
ie - a demand of 3 customers/hour divided by a capacity of 4 customers/hour = 75%
What is a methodology for planning resource allocation that is based on the management of flow unit queues? It focuses on the deployment of resources so that queue waits are acceptable
waiting line management, or queue management
Name 2 technical models for predicting performance
analytical queueing models and Monte Carlo simulations
Markovian
In these cases, both the time for arrival times and service times follow an exponential probability distribution.
In the standard coding system established for analytical queuing model categorization, the code is as follows: A/B/S/D/N/K.
What does each stand for?
A is the arrival time distribution
B is the service time distribution
S is the number of parallel servers
D is the queue discipline
N is the system capacity
K is the size of the population
Exponential arrivals with rate λ (lambda) is expressed as:
arrivals per time interval
Exponential service with rate μ (mu) is expressed as:
service completions per time interval
P0 is the probability that the server is idle. What is this formula?
P0 = 1−(λ / μ)
λ = arrivals per time interval
μ = service completions per time interval
Use the M/M/s model to determine the average wait time for customers to a motor vehicle driving test department, where customers arrive at a rate of 9.5 per hour and service times average 17 minutes per customer (exponentially distributed). Assume that 3 service providers are present.
Identify all the elements of the formula; this is done using excel sheet.
demand = 9.5cust/hr
service time = 17min/customer
capacity = 60min/hour divided by 17min/customer = 3.53cust/hour
utilization = demand/(capacity*n of servers)
= 9.5c/hr div by (3.53c/hr * 3servers)
=.871
The average waiting time would be 44.8 minutes.
If a target server utilization threshold of 85% is used, would this be considered a 15% buffer for the service system?
Yes, because it is guarding against the effects of demand variation on system responsiveness.
What is capacity?
If service time is 15 minutes, what is the capacity?
What does a person have the capacity to handle?
It is the rate that the items are made or customers are served.
It is the inverse of service time.
Example: if the average service time is 15 minutes
60min/hour divided by 15 min/customer
the capacity is 4 customers per hour
If there are 2 nurses both with a capacity of 4 customers/hour, then together they have a capacity of 2*4 = 8 customers/hour
When we have a Poisson arrival process, the time between arrivals is represented with:
an exponential distribution
What does M/M/s stand for?
M - arrival process is Markovian (the time between arrivals are exponentially distributed)
M - service time is Markovian (the time between arrivals are exponentially distributed)
s - number of servers in your system
At a call center where calls for troubleshooting are categorized by reason, why would the CV for total calls per day tend to be less than the CV for calls for each reason?
Because the total calls would combine the uncertainty across each reason.
What’s the recommended target effective capacity for:
1- reliable operating systems with little or not activity time variation
2- unreliable operating systems with highly variable activity times
1- 95%
2- 85%
On a graph that shows resource utilization and target utilization, we want the resource to utilization to be __ than the target utilization for a feasible capacity plan
less than (or equal to)
Monte Carlo simulations are: __ .
They are useful when the operating system: __
computer programs that mimic the operating system; it uses random number generators to mimic uncertainty within the simulation logic
is too complex for analytical modeling
The effectiveness of a queuing system is generally measured according to two perspectives; both perspectives are important to a decision maker. They include:
1 - from the customer’s POV – waiting time in queue or time in system
2 - from the server’s perspective – predicting server utilization
Service time variation is usually similar to the exponential distribution in that it tends to be skewed __.
Why would the variation in durations for a repair service be skewed ^ this way?
“right” (also known as a positive skew)
Because if the service is at all unusual or challenging, the repair time would typically be longer, rather than shorter, to the average repair time. That is, longer than average repair times would be more likely than shorter than average repair times.
