Distributed Architectures

Question 1

What are the two types of scaling?

Answer 1

Vertical and Horizontal

Question 2

What is Vertical Scaling?

Answer 2

Increasing the capacity or ability of a single system or machine

Question 3

What is Horizontal Scaling?

Answer 3

Adding more instances or server to distribute the workload across multiple machines

Question 4

What is replication?

Answer 4

A copy of the data is kept in each node

Question 5

What is partitioning?

Answer 5

Different nodes have different part of the data, each node stores a subset

Question 6

What are the advantages of replication?

Answer 6

1. Reducing latency as as data is geographically close
2. Increased availability
3. Increased Performance due to high No of machines to serve query

4.

Question 7

What are the disadvantages of replication?

Answer 7

1. The data has to be small enough to be stored on a single machine.
2. Write operations involve updating multiple copies of the same data
3. Updates need to be made on all nodes, if a write fails then data reliability can be affected

Question 8

How are nodes organised?

Answer 8

Nodes are followers with one node designated a leader. The leader has the most recent data and all write operations are sent to the leader.

Question 9

How are read and write operations handled with leader/follow dynamic?

Answer 9

All write operations are sent to the leader and once completed, send updates to the followers.

Read Queries can be sent to any node so are much quicker.

Question 10

How is Synchronous Replication used?

Answer 10

When the leader sends out a write request to followers only a subset of nodes need have confirmed completion. If all were required then a small issue could stop the whole system.

Question 11

What is a partitioned database?

Answer 11

A database that stores different subsets of the data in different nodes.

Question 12

How does partitioning store data?

Answer 12

A node can have more than one partition but each piece of data will belong to exactly one partitions.

Question 13

What is the goal of partitioning?

Answer 13

The goal is to spread the data and the query load evenly across nodes.

Question 14

What are Hotspot nodes?

Answer 14

Nodes that handle a disproportionately larger amount of data or load. Also known as skewed partitioning.

Question 15

What are three ways of partitioning data?

Answer 15

1. Randomly scatter the data across nodes
2. Partitioning by key range
3. Partitioning by Hash of key

Question 16

Why are partitions rebalanced?

Answer 16

To fairly share the load between the nodes in the cluster and reduce hot spot nodes.

Question 17

What re the two types of partitioning?

Answer 17

1. Fixed number
2. Dynamic

Question 18

How is fixed number partitioning implemented?

Answer 18

Many more partitions than nodes are created. If nodes are added, they can steal from the other nodes. Only entire partitions are moved between nodes, and the number of partitions doesn’t change, nor does the assignment of keys to partitions.

Question 19

How is dynamic partitioning implemented?

Answer 19

If partitions grow beyond a certain size they are split in two. If the data becomes smaller, partitions can be merged.

Question 20

What does CAP stand for?

Answer 20

Consistency, Availability, Partition Tolerance

Question 21

What is the CAP theorem?

Answer 21

In a distributed system only two of consistency, availability, and partition tolerance can exist together.

Question 22

What is system consistency?

Answer 22

Every read receives the most recent write or an error

Question 23

What is system availability?

Answer 23

Every request receives a non-error response.

Question 24

What is partition tolerance?

Answer 24

The system continues to operate even if the parts of the network become disconnected.

Question 25

What are the ACID properties?

Answer 25

Atomicity, Consistency, Isolation, Durability

Question 26

What is eventual consistency?

Answer 26

At a given time point some data nodes will have an outdated version of the data, overtime this will be updated resulting in eventual consistency and all the nodes have the same data.

Question 27

What is linearisability?

Answer 27

The idea that from a user perspective the system should appear as if all the operations on it are atomic and there is one single copy of data.

Question 28

How is linearizability achieved?

Answer 28

If one client reads a value then all subsequent clients should read the same value

Question 29

What are the characteristics of linearizability?

Answer 29

If applications require linearizability, some replicas are disconnected; applications that don't require linearizability can be more tolerant to network faults. It has performance impacts and isn't guaranteed in many systems in favour of performance.

Question 30

How do network delays affect response times?

Answer 30

If linearizability is desired, the response time of read and write requests should be at least proportional to the uncertainty of network delays.

Question 31

What is consensus?

Answer 31

Getting several nodes to agree on something.

Question 32

What is Atomicity?

Answer 32

The idea that a transaction that involves multiple nodes or services is treated as one indivisible piece of work. Either all commits are successful or none of the commits are.

Question 33

What is Consistency?

Answer 33

All nodes or replicas have the same view of data at a given time.

Question 34

What is a Two-phase commit?

Answer 34

An algorithm for achieving atomic transaction commits across multiple nodes