CAP theorem in System Design

CAP

• Consistency

• Availability

• Partition tolerance

Consistency

C -> Consistency

Consistency means the data should be consistent to all users for any application

Lets say A and B are the person who are experiencing a application,

but for any circumstances the application is visible only to B user then we can say that it is not consistent

Ex : A is depositing Rs 100 to the bank account and his balance is Rs 0 earlier, so after depositing rs 100, the balance should get updated.

It should not be 0 as it will lead to inconsistent system

Availability

A -> Availability

It refers that the system should be available all the time

We have often seen that when the SPPU results are out, the website is always down.

For a good system this should not happen

It has to be available all the time...

A good example can be Google.com (which is available all the time)

---

Partition Tolerance

P -> Partition tolerance

In a distributed system the application is divided into components and it is decentralized which means it is deployed on multiple servers not on the single server.

Lets assume one server gets down, we call it is as partition.

We call it is network partition AND

Tolerance means to save this server from getting down and if it gets down we will use the replica of that server

---

CAP Theorem

For a distributed system, the CAP theorem states that it is possible to attain only two properties from CAP and the third would always be compromised.

The requirement should define the two properties to choose from CAP properties.

As we can see, only two properties can be achieved at a time not all the three.

CP (Consistence and Partition tolerance)

Here we are compromising the availability of system

The system will be consistent and for this we will make the system unavailable for some time

and it is partition tolerance as if one of the node of server gets down, the replica of that node will take its place for functioning

Ex : If a bank application want to schedule a maintenance break, they will make the system unavailable for some time until the data is consistent

---

AP (Available and Partition tolerance)

Here the consistency of the system is compromised

The system should be available all the time

If we are writing a blog on a website it is not an emergency that the blog which is written by x-user needs to updated to all the readers at the same time. There can be a delay of 1 sec to the system to get consistent

Instead the availability of system is recommended.

System should be available.

---

CA (Consistent and available)

Here we say that the partition tolerance will be compromised

and if the Partition tolerance is skipped we no more have the distributed system

We will have the system as centralized system means there will be only one node available and that is the only way we can achieve consistence and availability.

And if the server is down, as we do not have partition tolerance the system will be down.

---

We have seen that for any system we can choose any 2 properties from CAP

But it is recommended that partition tolerance will always be there, so the choice comes to C and A i.e. Consistency or Availability

So the application will either be CP or AP

Let' s look at some examples and decide what to choose for a good system.

Ex :

1. Blog Website -> It needs to be available -> AP

2. Gaming platform -> AP

3. Stock trading -> The prices should be consistent across all the servers -> CP

4. Netflix -> AP

5. WhatsApp -> AP

6. Bank -> CP (we can make the system unavailable for some time as we cannot accept the inconsistent data)

We will choose only two options either CP or AP

This is all about CAP theorem in system design