Concepts and Technologies

Question 1

Load Balancing

Answer 1

• spread traffic across a cluster of servers
  to improve responsiveness and
  availability of services
• typically sits between client and server
• prevents a server from being a single
  point of failure
• keeps statuses on resources / health
  checks
• use active / passive cluster to prevent
  load balancer from being point of failure

BENEFITS:

• faster, uninterrupted service
• less server downtime = higher
  throughput
• fewer failed or stressed components
• predictive analytics = less bottlenecks

Ex. 1:
client –> internet –> load balancer –> servers

Ex. 2:
client –> LB –> web servers –> LB –> application servers –> LB –> databases

Question 2

Load Balancing Algorithms (How to pick which servers to send traffic to)

Answer 2

• fewest active connections
• lowest avg. response time
• least amount of traffic (bandwidth)
• round robin method (cycle through)
• weighted round robin (each server given
  a weight based on processing capacity,
  direct traffic to lowest weight first)

Question 3

Caching

Answer 3

• enable you to make better use of
  resources
• “recently requested data likely to be
  requested again”
• short term memory, limited space but
  faster
• can exist at all levels, often nearest to
  front-end

APPLICATION SERVER CACHE:
- placing a cache on the request layer
node enables the local storage of
response data
- request node –> check cache (fast) –>
otherwise go to disk

CONTENT DELIVERY NETWORK (CDN)
- for sites serving large amounts of static
media
- request to CDN –> serve or request to
back end servers –> cache it locally –>
client

Question 4

Cache Invalidation

Answer 4

• if data modified in db, needs to be
  invalidated in cache
• write-through: data written to both at
  same time. ensures nothing lost, but
  higher latency
• write-around: written to storage
  bypassing cache. smaller cache = higher
  latency
• write-back: only written to cache, then
  eventually storage. low latency and high
  throughput, but risk of data loss

Question 5

Cache Eviction Policies

Answer 5

• First in First Out
• Last in First Out
• LRU
• MRU
• LFU
• random replacement

Question 6

Relational Database (SQL)

Answer 6

• structured, with pre-defined schemas
• store data in rows/columns and the data
  needs to be compliant

Querying:
- uses SQL, which is very powerful

Scalability:
- vertically scalable, which is expensive.
- horizontal is challenging and time
consuming

Reliability:
- ACID compliant, better for data reliability

Use Cases:
- data reliability and ACID compliance
- if data is structured and unchanging
- no massive growth and consistent data
and traffic

Examples:
- Microsoft SQL Server
- Oracle SQL Server
- Postgre SQL Server

Question 7

Non-Relational Database (NoSQL)

Answer 7

• unstructured, dynamic schema. each
  row doesn’t have to have data for each
  column, can create columns on the fly.
• data is distributed

types:
- key-value stores
- document db: data stored in documents
in collections, documents can have
different structure

Querying:
- based on documents

Scalability:
- horizontally scalable, which is much
more cost effective

Reliability:
- sacrifice reliability for performance and
scalability

Use Cases:
- if all other components are fast /
seamless, db will not be the bottleneck
- storing large volumes of data with no
structure
- make the most of cloud computing and
storage. can be very cost effective, but
requires data to be easily spread across
servers to scale up.
- rapid development

Examples:
- Amazon DynamoDB
- MongoDB
- Cassandra