Method checkpoint: How to approach a design question

Method checkpoint

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 16 answered0 right on the first attempt

Question 1 of 16

A service expects 1 billion reads a month. Taking a month as 2.5 million seconds, how many reads a second is that on average?

Your answerRPS

Answers within 2% count.

Question 2 of 16

Each paste is about 1.27 KB and there are 10 million new pastes a month. How much new paste content is that per month?

Your answer

Answers within 2% count, in any unit you choose.

Question 3 of 16

You are asked to “design a photo-sharing service”. What should you do first?

Ask who uses it, how, how many users and requests, how much data, and the read to write ratioPick a CDN providerSketch the load balancer, web servers and databaseChoose a sharding key for the photos table

Question 4 of 16

Your estimate gives 40 reads a second on average, and you have assumed traffic is not evenly distributed. What does the primer’s Pastebin solution say about the peak?

That the average is enough to size the systemOnly that it is higher than the average, with no figureThat it is ten times the averageThat it is exactly twice the average

Question 5 of 16

You trace each use case through your sketch, and “view a paste” has no path to where the paste text is kept. What does that tell you?

Nothing: paths are detailed in step 3A cache is neededThe use case should be moved out of scopeA component or connection is missing from the sketch

Question 6 of 16

Put the primer’s four steps in order.

Design core components
Scale the design
Create a high level design
Outline use cases, constraints, and assumptions

Question 7 of 16

Why does the primer’s first Pastebin design have no cache and no read replicas?

They answer bottlenecks, which are identified in step 4Pastebin has more writes than readsThey would make pastes inconsistentThey are too expensive for a paste service

Question 8 of 16

Match each part of the first Pastebin design to the reason for it.

Object store for the paste textBatch job over web server logsSeparate Read and Write APIsSQL table keyed by short link

Question 9 of 16

Pastebin expects 360 million links over three years. Are 7 Base62 characters enough?

Only just: about 400 million valuesYes: 62 to the 7th is about 3.5 trillion valuesNo: 7 characters give only 62 × 7 = 434 valuesNo: Base64 is needed for that many

Question 10 of 16

Why does the primer’s solution encode the link in Base62 rather than Base64?

Base62 is faster to computeBase62 gives shorter links than Base64Base62 is random, so links cannot be guessedBase64’s + and / characters cause problems in URLs

Question 11 of 16

In the Pastebin design, what stops two pastes from getting the same short link?

Nothing: collisions are too rare to matterMD5 never produces the same output twiceThe short link is the primary key, and the Write API retries on a duplicateThe cache rejects repeated links

Question 12 of 16

You have ten minutes for step 3 of the Pastebin question. Where should you spend them?

A real-time analytics pipelineUser accounts and loginA paragraph on every component, equallyLink generation and the read path, saying that analytics stays a simple batch job

Question 13 of 16

Put the primer’s scaling loop in order.

Profile for bottlenecks
Address the bottleneck, weighing alternatives and trade-offs
Repeat
Benchmark or load test

Question 14 of 16

Pastebin averages 4 writes a second. What does the primer’s solution do about writes when scaling?

Puts every write through a queueShards the database from the startMoves all data to NoSQLKeeps one SQL primary and standby, which should manage 4 writes a second

Question 15 of 16

In the scaled Pastebin design, what handles reads of popular pastes, and what handles the misses?

A CDN for every pasteA memory cache for popular pastes; SQL read replicas for the missesDatabase shards, one per pasteThe SQL primary for everything

Question 16 of 16

An interviewer asks why you added read replicas. Which answer follows the method?

Large systems always have read replicasThey add write capacityWe might need them one dayCache misses were overloading the primary’s reads; replicas take them, at the cost of replication lag and replaying every write