Scaling

Question 1

What does DNS stand for?

Answer 1

Domain Name Service

Question 2

What does the DNS Server do?

Answer 2

Converts website domain names to IP Addresses

Question 3

What does HTTP stand for?

Answer 3

Hypertext Transfer Protocol

Question 4

What does IP stand for?

Answer 4

Internet Protocol

Question 5

What does JSON stand for?

Answer 5

JavaScript Object Notation

Question 6

Under what circumstances might you use a non-relational database over a relational database?

Answer 6

• you need super-low latency
• unstructured data
• You only need to serialize / deserialize data
• You need to store a large amount of data

Question 7

What are the 4 common types of non-relational databases?

Answer 7

1. key-value stores
2. graph stores
3. column stores
4. document stores

Question 8

What is a join operation?

For databases

Answer 8

Join combines data from multiple tables into a new dataset based on a specified condition (like a common field)

Question 9

Do non-relational databases support join operations?

Answer 9

Generally, no

Question 10

What is the difference between vertical scaling and horizontal scaling?

Answer 10

• Vertical Scaling involves making your existing resources more powerful
• Horizontal Scaling involves adding more resources to your pool of resources

Question 11

What are 2 reasons you would prefer horizontal scaling over vertical scaling

Answer 11

1. Vertical scaling has a hard limit
2. Vertical scaling does not have failover/redundancy.

Question 12

What does a load balancer do?

Answer 12

distributes incoming traffic among servers.

Question 13

In the context of Database Replication

Explain the difference of master DBs and slave DBs

Answer 13

• master DBs handle all write operations
• slave DBs handle all read operations

Question 14

What are 3 key benefits of database replication?

Answer 14

1. Performance (parallelization)
2. Reliability
3. High Availability

Question 15

Describe the data flow for a read-through cache

Answer 15

read(x):
- If x not in cache:
- cache[x] := db.read(x)
- return cache[x]

Question 16

Describe the data flow for a write-through cache

Answer 16

write(x, v):
- If x in cache:
- cache[x] := v
- db.write(x, v)

Question 17

Can a cache be both read-through AND write-through?

Answer 17

Yes

Question 18

Do you need a different cache in each datacenter?

Answer 18

Yes!

otherwise the cache can become a SPOF

Question 19

In cacheing

What is an expiration policy?

Answer 19

How long you wait before removing data from the cache

Question 20

What problem can happen if a cache expiration policy is too small on a cache?

Answer 20

You lose the speedup benefits of cacheing

Question 21

What problem can happen if a cache expiration policy is too large on a cache?

Answer 21

Data can become stale

Question 22

What does SPOF stand for?

Answer 22

Single Point Of Failure

Question 23

What is the best situation for cacheing?

Answer 23

When data is read frequently but modified infrequently

Question 24

What are the 5 key considerations around cacheing?

Answer 24

1. effectiveness (high read, low write)
2. Expiration Policy
3. Consistency
4. Failure Mitigation (avoid becoming SPOF)
5. Invalidation Policy

Question 25

What does **CDN** stand for?

Answer 25

**C**ontent **D**elivery **N**etwork

Question 26

What is a CDN?

Answer 26

A network of _geographically dispersed_ servers used to deliver _static content_

Question 27

What is the typical cost structure for a CDN?

Answer 27

charged *per data transfer in/out*

Question 28

When a user visits a website, which CDN server should deliver content to the user?

Answer 28

The CDN server _geographically closest to the user_

Question 29

Why is stateless architecture generally preferable to stateful architecture?

Answer 29

Because requests from the same client can be routed to different servers | without the overhead of sticky sessions ## Footnote So, better decoupling

Question 30

How might we be able to store "state" data in a stateless architecture?

Answer 30

Keep state data in a separate data store from the rest of the web layer architecture

Question 31

How might we improve website availability / performance across wider geographical areas?

Answer 31

Use multiple data centers (thing AWS regions / AZs), and geo-load balancing ## Footnote Using a CDN helps with geo-performance but *not availability*

Question 32

Describe a **message queue** architecture

Answer 32

(think SQS): X Producers putting work on the queue, Y consumers taking work off the queue; decouples producer and consumer work

Question 33

How can you perform horizontal scaling at the database level _without_ replicating your database?

Answer 33

**DB sharding** (e.g. with a shard hash of `primary_key % num_shards`)

Question 34

What are 3 downsides of sharding?

Answer 34

1. **Resharding** after growth 2. **Celebrity Problem** or an uneven traffic distribution 3. **Complicated Joins** (can be mitiigated by denormalizing data)

Question 35

What are the *8* big ideas for scaling?

Answer 35

1. Statelessness 2. Redundancy 3. Cacheing 4. Multiple Data Centers 5. CDNs for static assets 6. Sharding in Data Tier 7. Decoupling 8. Logging / Monitoring / Automation