Video summary
E-Business Capacity Planning (Cont.)
Main summary
Key takeaways
Summary
The lecture continues a discussion of e-business capacity planning: using customer behavior, workload forecasts, and performance models to determine whether a website’s infrastructure can meet expected demand.
Capacity-planning approach
- Break each website function into interactions between clients and servers.
- Represent computing resources as queues. When a site uses multiple resources, model them together as a queueing network.
- Use access logs and customer-behavior data to understand and forecast workload.
- Forecast demand for both the site overall and individual business functions, such as product searches or payments, since different functions may peak at different times.
- Use the forecast workload as input to performance models, then assess whether the current infrastructure can meet service expectations.
Example: Planning for higher business targets
The example site sells computer components, electronics, software, and gifts. Its revenue comes from merchandise sales and banner advertising. The lecturer cites historical merchandise revenue of $94 million and advertising revenue of about $0.9 million. About 97% of visitors make no purchase, and traffic during special sales periods may be many times higher than on an ordinary day.
Management’s goals for the next fiscal year are $130 million in merchandise revenue and $3 million in ad revenue. The example focuses on estimating the capacity needed for the merchandise target.
The calculation proceeds as follows:
-
Convert the revenue target to revenue throughput. Revenue throughput is the amount earned per second. The $130 million annual merchandise target corresponds to approximately $4.12 per second.
-
Estimate the required session rate. The lecture uses an average sale of $225 and a 4.6% buy-to-visit ratio. The relationship is:
Revenue throughput = sessions per second × buy-to-visit ratio × average sale
This gives a required rate of approximately 0.398 sessions per second.
-
Convert sessions into server requests. The average session contains 8.144 page requests. Multiplying this by the required session rate gives a typical expected workload of about 3.241 requests per second.
-
Allow for peak traffic. Based on past observations, the lecture assumes a major selling-period burst can reach 20 times the normal load. Under that assumption, the site may need to handle approximately 64.82 requests per second.
Assessing and responding to capacity needs
- Treat the forecast request rate as the arrival rate (λ) in the queueing model.
- Estimate the server’s service rate (μ) from access-log data using observed service times.
- Use the model to calculate performance measures such as waiting time and server utilization.
- Check whether the resulting service quality meets management’s expectations. Utilization approaching 1 indicates that a server is near saturation; the lecturer suggests that utilization above roughly 0.7 may warrant concern.
- If the infrastructure is insufficient, consider:
- Scaling up: increasing the capacity of existing servers.
- Scaling out: adding more servers, such as in a cluster.
- Because this example concerns demand expected over the coming year, rather than just a brief event, capacity changes should be planned as a longer-term infrastructure decision.
Speakers and sources
- Speaker: One unnamed lecturer or narrator.
- Sources and inputs discussed: Website access logs, customer-behavior and workload models, historical sales data, management’s revenue targets, and queueing-model concepts.
Rate this summary
Your feedback will help improve summaries.
Improve this summary
Reprocess with a stronger model when the summary feels incomplete or inaccurate.
Translate summary in another language
Ask questions to this video
Chat for follow-up questions, clarifications, and source-backed answers.