Skip to main content

CAP theorem demystified


In a web world, everything is available over the internet. While designing a distributed application for the client, we often face this dilemma of choosing between consistency and availability.  There is no further argument if we don't choose to have a partition tolerant system, that's single node application and it will adhere to consistency and availability. So we will be discussing Partition tolerant systems for a distributed environment.  Lets first understand CAP individual elements;

C (Consistency): 
In a distributed environment of n nodes, a change in one node should be instantly reflected all other nodes. So any data change/state should be reflected all client irrespective of whichever node they access the data. 


A (Availability):

All the non-failing node should be ready to serve any request within a reasonable time. This applies to every node present in the system. 

P (Partition Tolerence):

Distributed systems are meant to build partition tolerant systems. Any node creation and removal should be gracefully handled. 


A distributed system need to partition tolerant, So P axis will be fixed. As there is a contradiction between Consistency and availability, we have to fine-tune them for business cases. 

Example: I have 3 servers with user data in place. Now All three have the same data, so any read request will be having consistent data for any read request to any node.  Now if user A has modified his details and call has been made to server S1 at T0. So data is updated at S1 at T0 but will be propagated to S2 and S3 post T4 (Assume). So either S2 and S3 will choose to decline any request till T4 to maintain consistency and compromise availability or will continue to serve request but with an outdated data. 

And that's gist of CAP theorem. 

Comments

Post a Comment

Popular posts from this blog

Car Parking Problem

There is n parking slots and n-1 car already parked. Lets say car parked with initial arrangement and we want to make the car to be parked to some other arrangement. Lets say n = 5, inital = free, 3, 4, 1, 2 desired = 1, free, 2, 4 ,3 Give an algorithm with minimum steps needed to get desired arrangement. Told by one of my friend and after a lot of search i really got a nice solution. I will post solution in comment part
3 comments
Read more

Median of Five Numbers

U have 5 NOs , X1,X2,X3,X4,X5 With minimum no. of comparisons we have to find a median. SWAP(X,Y) function is available to u . I have a answer of six comparisons and eight swaps....wait for people to find out by themselves.
1 comment
Read more

Consistent Hashing

I will try to explain consistent hashing with a real world example. Let's assume I have a restaurant with 60 tables and 5 servers (waiter). Each server is given an equal number of tables to serve. Now let's assume we have addition of a new server (waiter), so his addition will be marked in the circle and he will be receiving tables from the previous server to his distance only. Check the attached example. Assume a server (waiter) has left the organisation and we have only 4 servers now. Server3 has left the restaurant, so his table will be assigned to server 4. I am sure you have noticed the load is not equally distributed. But to make the system less prone to addition/removal we just rotate in clockwise and assign range from the previous server to present server.  To make sure load is balanced or optimally balanced we need to use virtual nodes. Check links here: http://tom-e-white.com/2007/11/consistent-hashing.html https://www.toptal.com/big-data/...
Post a Comment
Read more