Junior Scalability & Performance Interview Questions (50 MCQs)

Question 1

What is meant by vertical scaling (scale up)?

Accepted Answer

Adding more resources such as CPU, RAM, or disk to one server

Answer

Adding more servers behind a load balancer

Answer

Splitting data across several independent databases

Answer

Moving the service into several geographic regions

Question 2

What is meant by horizontal scaling (scale out)?

Accepted Answer

Adding more machines and distributing load among them

Answer

Upgrading the current server's CPU to a faster model

Answer

Increasing server RAM to hold a bigger cache

Answer

Optimizing queries so they use less CPU

Question 3

What is the main limitation of vertical scaling?

Accepted Answer

It hits a hardware ceiling and one machine is still a single point of failure

Answer

It requires rewriting the whole codebase to run on many nodes

Answer

It is not possible at all without a load balancer

Answer

It always ends up cheaper than horizontal scaling

Question 4

When a web service is scaled horizontally, what component is usually placed in front of the servers?

Accepted Answer

A load balancer that distributes requests across instances

Answer

A message broker that stores every response

Answer

A central database that routes the requests

Answer

A cron job that turns servers on in rotation

Question 5

What does it mean for a service to be stateless?

Accepted Answer

It keeps no data from a user's previous requests in its own memory

Answer

It does not use any database or external storage

Answer

It only responds to `GET` requests

Answer

It restarts its process after every request

Question 6

To keep app servers stateless, where are user sessions usually stored?

Accepted Answer

In a shared store like Redis or inside a signed token

Answer

In the memory of whichever app server the user hit first

Answer

In local files on each app server's disk

Answer

In global variables inside each instance's code

Question 7

What is the main drawback of sticky sessions (pinning each user to one server)?

Accepted Answer

Load becomes uneven and the user's session is lost if that server fails

Answer

It causes every request to be processed twice

Answer

It makes using HTTPS impossible

Answer

It forces the database to lock all tables

Question 8

What does latency mean in a system?

Accepted Answer

The time it takes for a single request to get a response

Answer

The number of requests the system processes per second

Answer

The maximum amount of data the network moves per second

Answer

The percentage of time the system is available

Question 9

How is the throughput of a service usually expressed?

Accepted Answer

Requests or units of work completed per unit time, like RPS

Answer

The average response time of each request in milliseconds

Answer

The number of servers behind the load balancer

Answer

The total volume of data stored in the database

Question 10

Adding more servers behind a load balancer most directly improves which metric?

Accepted Answer

Total system throughput, not necessarily per-request latency

Answer

Per-request latency, even when the system is lightly loaded

Answer

The execution speed of slow database queries

Answer

The network round-trip time between user and data center

Question 11

If an API's p50 latency is 80 ms, what does that mean?

Accepted Answer

Half of the requests completed in 80 ms or less

Answer

The arithmetic mean of all requests is exactly 80 ms

Answer

50 percent of the requests failed with an error

Answer

The slowest request took 80 ms

Question 12

What does a p99 latency of 1.2 seconds indicate?

Accepted Answer

99% of requests finished within 1.2 s and 1% were slower

Answer

99% of requests took exactly 1.2 seconds

Answer

The system was available 99% of the time at 1.2 s latency

Answer

The average of the last 99 requests was 1.2 seconds

Question 13

Why is reporting only the average latency not enough for monitoring?

Accepted Answer

The average hides very slow requests and does not show the distribution

Answer

Computing an average is mathematically impossible for large systems

Answer

The average is always larger than p99 and causes false alerts

Answer

The average is only defined for throughput, not latency

Question 14

What does the term tail latency refer to?

Accepted Answer

The latency of the slowest fraction of requests, like p99 or p99.9

Answer

The latency of the last request served before server shutdown

Answer

The time it takes to append logs to the end of a file

Answer

The delay between sending the last response byte and closing the connection

Question 15

In capacity estimation, what does QPS stand for and what does it measure?

Accepted Answer

Queries Per Second; the number of requests arriving at the system each second

Answer

Quality Per Service; the response quality of each service

Answer

Queue Processing Speed; how fast a message queue drains

Answer

Query Payload Size; the average size of each request

Question 16

In back-of-the-envelope math, roughly how many seconds are in a day?

Accepted Answer

About 86,400, usually rounded to `10^5`

Answer

About 8,640, usually rounded to `10^4`

Answer

About 864,000, usually rounded to `10^6`

Answer

About 3,600, usually rounded to `10^3`

Question 17

A service has 1 million daily active users, each sending 10 requests per day. What is the approximate average QPS?

Accepted Answer

About 120 requests per second

Answer

About 1,200 requests per second

Answer

About 12 requests per second

Answer

About 10,000 requests per second

Question 18

Why do we not size a system based only on its average QPS?

Accepted Answer

Because peak-hour traffic is several times the average and the system must handle it

Answer

Because average QPS is always higher than peak QPS

Answer

Because average QPS can only be computed after launch

Answer

Because load balancers only work with peak QPS

Question 19

Which is the correct order from fastest to slowest for reading data?

Accepted Answer

CPU cache, RAM, SSD, spinning hard disk

Answer

RAM, CPU cache, spinning hard disk, SSD

Answer

SSD, RAM, CPU cache, spinning hard disk

Answer

CPU cache, SSD, RAM, spinning hard disk

Question 20

Roughly what is the network round-trip time (RTT) between two continents, such as Europe and the US?

Accepted Answer

About 100 to 150 milliseconds

Answer

About 1 to 2 microseconds

Answer

About 0.5 milliseconds

Answer

About 5 to 10 seconds

Question 21

What is a bottleneck in a system?

Accepted Answer

The component with the lowest capacity that limits overall throughput

Answer

The fastest component that all others must keep up with

Answer

Any server that sits behind a load balancer

Answer

A part of the code that never gets executed

Question 22

What is the main purpose of database connection pooling?

Accepted Answer

Reusing open connections instead of creating a new one per request

Answer

Storing query results to answer repeated queries faster

Answer

Splitting a table's data across several database servers

Answer

Automatically encrypting all database connections

Question 23

Why do we hand off slow tasks like sending email or processing video to a queue and background workers?

Accepted Answer

So the user gets a fast response and heavy work scales separately

Answer

So the heavy work runs faster on the same request thread

Answer

So no database is needed anymore to store results

Answer

So the execution order of all user requests is guaranteed

Question 24

What is autoscaling?

Accepted Answer

Automatically adding or removing instances based on metrics like CPU or load

Answer

Automatically upgrading the OS version on all servers

Answer

Automatically compressing responses when traffic is high

Answer

Automatically changing the database schema as data grows

Question 25

What is an SLI (Service Level Indicator)?

Accepted Answer

A measured metric of service behavior, like the percentage of successful requests

Answer

The team's internal quality target for the service, like 99.9%

Answer

A legal contract with the customer including penalties

Answer

The list of servers the service runs on

Question 26

Which option is an SLO (Service Level Objective)?

Accepted Answer

99.9% of requests must succeed over a 30-day window

Answer

The number of failed requests in the last minute

Answer

If uptime drops below 99%, 10% of the fee is refunded to the customer

Answer

The service runs on 12 Kubernetes pods

Question 27

What is the main difference between an SLA and an SLO?

Accepted Answer

An SLA is a customer contract with financial or legal consequences for breach

Answer

An SLA is always set stricter than the team's internal SLO

Answer

An SLA only measures latency and an SLO only availability

Answer

An SLA is the measurement and an SLO is the customer contract

Question 28

Roughly how much downtime per 30-day month does 99.9% (three nines) availability allow?

Accepted Answer

About 43 minutes

Answer

About 4.3 minutes

Answer

About 7.2 hours

Answer

About 26 seconds

Question 29

For a read-heavy system like a popular blog, which approach is usually the first and most effective?

Accepted Answer

Adding a cache and read replicas to spread read load

Answer

Indexing every column to make writes faster

Answer

Removing the database entirely and storing everything in files

Answer

Processing all reads as a nightly batch job

Question 30

A system that writes tens of thousands of log or metric events per second has what kind of workload?

Accepted Answer

Write-heavy; so batching and write-optimized storage matter

Answer

Read-heavy; so adding a cache is the key optimization

Answer

CPU-bound; so only a faster CPU solves it

Answer

Stateless; so it needs no durable storage at all

Question 31

What does denormalization mean in a database?

Accepted Answer

Deliberately duplicating data so reads avoid expensive joins

Answer

Removing all indexes to shrink the database

Answer

Splitting every table into smaller ones to remove duplication

Answer

Encrypting sensitive columns for better security

Question 32

Why is inserting 1,000 rows in one batch query usually faster than 1,000 separate queries?

Accepted Answer

Per-request fixed overhead like network round trips and commits is paid only once

Answer

The database disables constraint validation in batch mode

Answer

A batch query is automatically spread across several database servers

Answer

In batch mode data is stored without writing to disk

Question 33

What trade-off does enabling compression (like gzip) on HTTP responses involve?

Accepted Answer

More CPU usage in exchange for less transfer size and bandwidth

Answer

Larger transfer size in exchange for less CPU usage

Answer

Less security in exchange for higher latency

Answer

Less memory usage in exchange for losing some data

Question 34

Why should an API not return all rows of a large table in a single response?

Accepted Answer

It wastes memory, processing time, and bandwidth; pagination is needed

Answer

Because HTTP does not allow responses larger than 64 KB

Answer

Because JSON cannot hold arrays with more than 100 elements

Answer

Because databases can only return 10 rows per query

Question 35

What is the effect of adding an index to a database column?

Accepted Answer

Filtered reads on that column get faster and writes get slightly slower

Answer

Both reads and writes get faster with no cost at all

Answer

Writes get faster and reads get slower

Answer

The database gets smaller because data is compressed

Question 36

What is the N+1 query problem?

Accepted Answer

One query fetches a list, then a separate query runs for each item

Answer

The database opens one extra connection for every query

Answer

Every table needs one extra column for its id

Answer

Queries always return one more row than needed

Question 37

Why does putting an in-memory cache like Redis in front of a database reduce latency?

Accepted Answer

Hot data is read from memory, skipping database queries and disk

Answer

Redis executes SQL queries faster than the database itself

Answer

The cache compresses data and makes it smaller

Answer

The cache limits the number of concurrent users

Question 38

Why do large companies usually prefer horizontal scaling on commodity machines over one giant server?

Accepted Answer

Cost per unit of capacity is lower and one machine failing does not take down the service

Answer

Commodity machines process each request faster than a giant server

Answer

Horizontal scaling needs no network between servers

Answer

With horizontal scaling monitoring is no longer needed

Question 39

An app saves user uploads on the app server's local disk. What problem appears when scaling out to multiple servers?

Accepted Answer

Requests hitting another server cannot find the file; shared object storage should be used

Answer

Files are automatically copied to every server and disks fill up

Answer

The load balancer does not allow file uploads

Answer

Upload speed is divided by the number of servers

Question 40

How does network bandwidth differ from network latency?

Accepted Answer

Bandwidth is maximum data per unit time; latency is how long one packet takes to arrive

Answer

Bandwidth is how long one packet takes; latency is data volume per second

Answer

They are the same concept with different units

Answer

Bandwidth only applies to wireless and latency only to fiber

Question 41

If we store a 1 KB profile for each of 10 million users, what is the approximate size?

Accepted Answer

About 10 GB

Answer

About 10 MB

Answer

About 1 TB

Answer

About 100 GB

Question 42

In back-of-the-envelope math, 1 terabyte is roughly how many bytes?

Accepted Answer

`10^12` bytes

Answer

`10^9` bytes

Answer

`10^15` bytes

Answer

`10^6` bytes

Question 43

Why should every network call to another service or database have a timeout?

Accepted Answer

So a slow dependency cannot hold threads and connections forever

Answer

So requests always finish in exactly a fixed time

Answer

So retries are never needed under any condition

Answer

So the network bandwidth usage is reduced

Question 44

Before optimizing a service's performance, what is the best first step?

Accepted Answer

Measure and profile to find the real bottleneck

Answer

Rewrite the whole service in a faster language

Answer

Double the number of servers just to be safe

Answer

Add a cache in front of every endpoint

Question 45

What is the purpose of load testing a system?

Accepted Answer

To learn how the system behaves and its max capacity under realistic or higher load

Answer

To verify business logic correctness in individual functions

Answer

To find spelling mistakes in the user interface

Answer

To reduce server costs in production

Question 46

Why is scaling out the database tier usually harder than scaling out app servers?

Accepted Answer

The database is stateful and spreading data across nodes raises consistency and coordination issues

Answer

Databases can only run on a single CPU core

Answer

Load balancers do not support database protocols

Answer

Database license cost multiplies tenfold with each node

Question 47

A service spends most of its time waiting on database and external API responses. What kind of service is it?

Accepted Answer

I/O-bound; a faster CPU will not help it much

Answer

CPU-bound; upgrading the CPU directly speeds it up

Answer

Memory-bound; only more RAM will speed it up

Answer

Stateless; so it will never have a bottleneck

Question 48

How does rate limiting help an API's scalability?

Accepted Answer

It stops one client from overusing resources and keeps the service stable for others

Answer

It cuts the latency of every request in half

Answer

It automatically increases database capacity

Answer

It removes the need for a load balancer

Question 49

Why is `SELECT *` on large tables discouraged in production code?

Accepted Answer

It reads unneeded columns and wastes I/O, memory, and network transfer

Answer

Because `SELECT *` always locks the entire table

Answer

Because `SELECT *` returns results in no order

Answer

Because `SELECT *` is not supported in modern databases

Question 50

Why serve static files like images and JS from a CDN rather than the app server?

Accepted Answer

A CDN caches files near users and takes load off the app server

Answer

A CDN executes JavaScript faster than the browser

Answer

App servers are unable to serve binary files

Answer

A CDN removes the need for a database for the whole site