Chapter 1 checkpoint

Chapter 1 checkpoint: system design foundations

Every question from this chapter’s lessons. Your first answer to each is the one that counts; afterward you can practice again as often as you like. Passing needs 75% right on the first attempt.

0 of 17 answered0 right on the first attempt

Question 1 of 17

A product page takes four seconds to load even late at night, when nobody else is using the site. What kind of problem is this?

A consistency problemAn availability problemA scalability problemA performance problem

Question 2 of 17

Four identical web servers together handle 2,000 requests per second. If the service scales perfectly in Vogels’ sense, how many requests per second will ten of them handle?

Your answerRPS

Answers within 2% count.

Question 3 of 17

Your web servers keep each user’s session in their own memory. What has to change before you can run several of them behind a load balancer?

Keep sessions in a shared store such as Redis, or in signed client-side tokensGive each server more memory so it can hold more sessionsRun a copy of the database on every web serverTurn off the load balancer’s health checks

Question 4 of 17

You double the number of web servers, but total throughput does not change and the database’s CPU sits at 100%. What is the most likely explanation?

Horizontal scaling never increases throughputThe database is now the bottleneck, so more web servers cannot add throughputA network partition is splitting the web serversThe web servers’ code is too slow

Question 5 of 17

An API receives 5,000 requests per second, and each request spends 40 ms in the system. On average, how many requests are in flight at once?

Your answerrequests

Answers within 5% count.

Question 6 of 17

In the M/M/1 model, a server needs 10 ms of work per request and runs at 90% utilization. What is the average time a request spends in the system?

Your answer

Answers within 5% count, in any unit you choose.

Question 7 of 17

A page load calls 40 backend services in parallel, and each is independently slow 1% of the time. What percentage of page loads hit at least one slow response?

Your answer%

Answers within 5% count.

Question 8 of 17

Match each term to its meaning.

LatencyThroughputp99 latencyUtilization

\rho

Question 9 of 17

In the CAP theorem, what does consistency mean?

Every read receives the most recent write or an errorEvery transaction leaves the database satisfying its constraintsEvery request receives a non-error responseAll replicas agree eventually, once writes stop

Question 10 of 17

A network partition separates two groups of Cassandra nodes running with default settings. What happens while it lasts?

The smaller side refuses requests so that no read is staleThe whole cluster shuts down to avoid losing dataBoth sides keep serving reads and writes, and their data can diverge until the partition healsSynchronous locks prevent the partition from happening

Question 11 of 17

Match each system to how the lesson classifies it.

etcdCassandra (default settings)Google Spanner

Question 12 of 17

PACELC adds a trade-off that applies when there is no partition. What is it?

Cost against durabilityAvailability against partition toleranceThere is no trade-off without a partitionLatency against consistency: waiting for replicas adds delay

Question 13 of 17

Put the stages of the primer’s walkthrough in the order it introduces them.

Add a load balancer, redundant web servers and a CDN
Add autoscaling
Add a cache and read replicas
Move the database and static content off the web server
Run everything on a single box
Move slow work to queues, and data to NoSQL and a warehouse

Question 14 of 17

A catalog service is 85% reads and 15% writes, and its single primary database is at 95% CPU. What is the most cost-effective first step?

Shard the database across ten clustersRewrite the application in a faster languageAdd a cache and read replicas for the readsMove the primary to the largest instance available

Question 15 of 17

Sending a confirmation email takes two seconds, and the sign-up request waits for it. What should the sign-up endpoint do instead?

Raise the load balancer’s timeoutCache the email in RedisPut a job on a queue for a worker and respond immediately, for example with 202 AcceptedAdd more web servers

Question 16 of 17

A payment service runs one load balancer, four application servers and one primary database. Which parts would take the whole service down if a single machine failed?

Only the databaseThe application serversThe load balancer and the databaseNone of them, because there are four application servers

Question 17 of 17

A service takes 6,000 requests a second, 90% of them reads. Reads try the cache first, which answers 80% of them; misses and all writes go to the database. Which component saturates first?

Load balancerHandles up to 50,000 requests/s

App servers ×4Each handles 2,000 requests/s

CacheHandles up to 100,000 lookups/sPrimary databaseHandles up to 1,500 queries/s