System Design and Analysis Quiz

15 questions on feasibility analysis and the system-design building blocks: scale, traffic routing, data, asynchronous work, resilience, and operations.

Analysis and feasibility

#

A proposed system has a positive financial return, but depends on an unfamiliar platform and must integrate with several legacy vendor systems. Which conclusion is most useful?

It may be economically feasible, but its technical feasibility is risky and needs investigation, prototypes, or mitigation

Correct. Economic and technical feasibility answer different questions. Familiarity and compatibility are technical-risk drivers even when the projected return is attractive.

It is feasible because a positive financial return settles every feasibility question

It is infeasible because legacy integrations always fail

Its only remaining concern is whether users like the interface

#

Which cost is easiest to miss when estimating the lifetime cost of a new hosted application?

Recurring operating costs such as support, maintenance, hosting, and licence renewals

Correct. Development is a one-off cost; operating costs can continue for years and must be included in the cash flow.

The initial development salaries

Hardware bought specifically for the first release

A consultant hired for the migration

#

Why should a feasibility study record an important benefit even when it cannot be reliably expressed in money?

It may be a real intangible benefit, such as better customer service, and omitting it would distort the decision

Correct. Intangible benefits should be estimated where possible and explicitly listed where they are not, rather than silently treated as worth zero.

Only monetary benefits can affect a project decision

It allows the project team to count the benefit twice in ROI

It removes the need to estimate costs

#

A project has low technical risk and strong strategic alignment, but the frontline users oppose the redesigned workflow. What should happen next?

Treat organisational feasibility as a material risk and plan user involvement, training, a phased rollout, or workflow changes

Correct. Adoption is part of feasibility. Resistance is information for the plan, not a reason to ignore the affected users.

Proceed unchanged because technical feasibility is more important than adoption

Remove users from the stakeholder list so their resistance cannot delay approval

Recalculate only the database capacity estimate

#

What does a negative net present value (NPV) mean for a proposed project at the chosen discount rate?

Its discounted future benefits do not cover its discounted costs, so the investment is not justified on that financial basis

Correct. NPV compares cash flows in present-value terms; a negative result means the money has a better use under the stated assumptions.

The project will never have any revenue

The project has no technical risks

The project breaks even immediately

Designing for scale and traffic

#

During a design review, what should be established before choosing a database, cache, or queue?

The goals, main features, expected traffic and data scale, read/write ratio, latency needs, and important constraints

Correct. Requirements and rough estimates give component choices a purpose; the architecture should grow from pressures rather than a technology checklist.

The cloud provider's most popular managed services

The exact number of microservices the team will deploy

A complete list of every possible failure mode

#

An API has many requests that take widely different amounts of time. Which load-balancing approach most directly accounts for that difference?

Least connections, which sends new work to the server with the fewest active connections

Correct. It is useful when long-running requests make a simple request count a poor approximation of current load.

Round robin, because it always selects the fastest server

IP hash, because it measures each request's execution time

Sticky sessions, because they distribute load evenly

#

Why should application servers generally avoid keeping user session state only in local memory when they sit behind a load balancer?

Any healthy server should be able to serve a request; a shared session store supports this and avoids dependence on sticky routing

Correct. Server-local sessions make scaling and failover harder because subsequent requests may reach a different server.

Local memory is slower than a network call

A load balancer cannot forward HTTP requests

Shared session stores prevent all application errors

#

A service must route /images to a static-content service and /api/orders to an orders service. What capability is needed?

Layer 7 load balancing, because it can inspect HTTP paths and other application-level information

Correct. Layer 4 routing uses transport information such as IP addresses and ports; path-based routing needs Layer 7 awareness.

Layer 4 load balancing, because paths are part of TCP

Database replication, because it routes requests by URL

A dead-letter queue, because it stores unmatched requests

Data, asynchronous work, and operations

#

An application uses cache-aside for a product catalogue. What is the normal sequence after a cache miss?

The application reads the database, returns the value, and stores it in the cache for later requests

Correct. With cache-aside, the application owns the loading logic and populates the cache after a miss.

The database reads from the cache and deletes the source record

The cache accepts the write and never uses durable storage

The load balancer retries the request on every server

#

Thousands of cache keys expire at exactly the same moment and the database suddenly receives all the resulting reads. What is this called, and what is one defence?

A cache stampede; add TTL jitter or coalesce refresh requests so only one request reloads a key

Correct. Coordinating refreshes, staggering expiration, or serving briefly stale data prevents simultaneous misses from overwhelming the source.

Backpressure; add sticky sessions

Write-through; add a read replica

A dead-letter queue; acknowledge every cache hit

#

A customer places an order, but sending the confirmation email should not delay the order response. Which design is most appropriate?

Save the order and publish an event or message for a background email worker

Correct. A queue decouples slow, non-critical work from the request path while letting workers process it asynchronously.

Call the email service synchronously and wait indefinitely for success

Store the email only in the application server's memory

Route email through a read replica

#

Why must consumers usually be idempotent when a queue provides at-least-once delivery?

A message can be delivered again after a crash or missing acknowledgement, so repeated processing must not produce an incorrect result

Correct. At-least-once delivery reduces message loss by allowing duplicates; the consumer must make duplicates safe.

Each message is guaranteed to be processed exactly once

Idempotence makes acknowledgements unnecessary

It ensures messages are received in global order

#

What is the main purpose of a dead-letter queue?

Keep repeatedly failing messages for inspection or special handling without blocking the main queue

Correct. After bounded retries, moving a poison message aside protects normal throughput and leaves evidence for diagnosis.

Copy every successful message for long-term analytics

Ensure every consumer receives each message

Replace health checks for failed servers

#

Which observation most strongly indicates that a system's consumers cannot keep up with producers?

Queue depth and message delay keep increasing over time

Correct. A growing backlog is backpressure: work arrives faster than it is completed. Add capacity, slow producers, or split the workload after finding the bottleneck.

The cache hit rate is high

A health check returns 200

The primary database has a read replica

Quiz Progress

• Answered: 0 / 0 (0%)
• Correct: 0 / 0 (0%) Reset