Designing Systems that Scale

Question 1

What are the pros and cons of the single-server design?

Answer 1

Pros: Cheap, easy to maintain
Cons: Single point of failure, cannot scale to support more users

Question 2

In a single-server design, what is the main benefit of separating the database from the main server into a separate resource?

Answer 2

You can now scale the server and database independently

Question 3

What is vertical scaling?

Answer 3

Replace your current webserver with a larger, more expensive webserver with increased cpu and memory

Question 4

What are the pros and cons of vertical scaling?

Answer 4

Pros: easier to maintain
Cons: Expensive, single point of failure, limited scalability as you get larger

Question 5

What is the purpose of a load balancer in horizontal scaling?

Answer 5

1. It distributes work evenly across multiple machines
2. It can reroute traffic to different servers if one or more goes down

Question 6

What is horizontal scaling?

Answer 6

When you scale your application with a cluster of machines which are given a distributed amount of work by a load balancer

Question 7

What are the pros and cons of horizontal scaling?

Answer 7

Pros: Resilient, infinitely scalable
Cons: More instances to manage, individual web servers must be stateless

Question 8

In the horizontal scaling technique, what does it mean that the individual web servers must be stateless?

Answer 8

The web servers have to be stateless because any request from the client must be able to be sent to any web server in the cluster. Can’t assume that one web server knows the state of a previous request and can act on it.

Question 9

T/F: Choose the simplest architecture that meets your projected traffic requirements, but no simpler

Answer 9

True

Question 10

Explain the ‘cold standby’ failover strategy

Answer 10

If a database goes down, you use periodic backups to restore your data on another database/server

Question 11

Explain the pros/cons of the ‘cold standby’ failover strategy

Answer 11

Pros: It’s better than nothing, prevents you from losing all your data. Cheaper option and simple.
Cons: Takes a long time to transfer backup data to new database, you’ll lose any data that wasn’t backed up in the last periodic backup

Question 12

Explain the ‘warm standby’ failover strategy

Answer 12

You have a mirror image of your database that can be instantly ready to take the place of the main database (replication)

Question 13

Explain the pros/cons of the ‘warm standby’ failover strategy

Answer 13

Pros: Near instant switch to backup, little/no data loss, easy to turn on (handled by db platforms)
Cons: More expensive, tiny chance that you might still lose data during the switch to backup

Question 14

Explain the ‘hot standby’ failover strategy

Answer 14

Front end webserver is writing the same data to multiple backup hosts at the same time, so if one host goes down the traffic is immediately routed to another host.

Question 15

Explain the pros/cons of the ‘hot standby’ failover strategy

Answer 15

Pros: most resilient, essentially a horizontal scaling solution
Cons: Most expensive

Question 16

What is a shard in a horizontally scaled database?

Answer 16

It’s a horizontal partition of your database

Question 17

What are the pros/cons of sharding your database?

Answer 17

Pros: It allows ultimate resiliency by backing up specific pieces of your data. Scalable as your database grows
Cons: It can get complicated joining/merging shards back together, so you need to be aware of how to best organize your data across partitions to minimize them

Question 18

What is MongoDB?

Answer 18

It’s a distributed NoSQL database which stores and partitions data based on key-value pairs, and manages replication

Question 19

What is Amazon RDS?

Answer 19

It’s a service in AWS that manages relational database servers and allows you to quickly query structured data

Question 20

What is Amazon Redshift?

Answer 20

It’s a service in AWS that supports data warehouse/lake approaches, enabling it to access and analyze huge amounts of data

Question 21

What are the scalability differences between RDS and Redshift?

Answer 21

• RDS can only scale vertically, but can generate multiple read replicas to support increased user traffic.
• Redshift can scale vertically and horizontally due to the nature of how it distributes data across nodes

Question 22

What are the tradeoffs between RDS and Redshift?

Answer 22

RDS delivers fast query searches and is cheaper, however it can only scale vertically which is a limitation if you need to deal with lots of data. Redshift can scale horizontally, allowing it to store massive amounts of data. However, queries can take much longer due to the time it takes to query multiple nodes.

Question 23

What does it mean that SQL databases follow a relational model?

Answer 23

It means that structured data is stored in tables, and foreign keys are used to form relationships across multiple tables

Question 24

What does it mean that NoSQL databases follow a non-relation model?

Answer 24

It means that data is stored as documents, key-value pairs, or graphs where there is no defined schema

Question 25

What’s another name for NoSQL databases?

Answer 25

Sharded databases

Question 26

What is the “hot spot” or “celebrity” problem in NoSQL databases?

Answer 26

A “hot spot” is a shard that gets much more traffic than any other shard for whatever reason. Because of this, you need to be aware/scale for this type of increased traffic.

Question 27

What are the main traits of normalized data?

Answer 27

Less storage space, more lookups (because data is stored in separate tables), updates happen in one place (one table)

Question 28

What are the main traits of denormalized data?

Answer 28

More storage space (because data is often redundant to keep relevant data close together), one lookup (assuming you organized relevant data together), updates are difficult (because you have to update multiple instances of the same data)

Question 29

What is a big determining factor on whether or not you denormalize your data?

Answer 29

It largely depends on the types of calls you plan on making to the data for the overall customer experience

Question 30

What is AWS’s ‘data lake’ solution?

Answer 30

S3

Question 31

T/F: data in a data lake has to be structured

Answer 31

False, a data lake is just a giant pile of unstructured data