20 Important System Design Concepts
Note: This list is taken from the Neetcode's Video 20 System Design Concepts Explained in 10 Minutes. Do watch the video, its highly informative!
Imagine, we have a single server that we have to scale to a large number of Users, we have to two methods to scale our Backend Server :-
1. Vertical Scaling
Vertical Scaling is to increase the resources on the same server i.e increasing the RAM, upgrading the CPU, adding GPU for more faster processing etc. So, in general, Vertical scaling means, to increase the capacity of the backend servers that we already own or improve the servers!.
NOTE: In this type of scaling, where we only have one server, if that server goes down or fails, then our complete system will fail! This Problem is called Single Point of Failure. To avoid this problem, usually system always have more than one servers running, so if one server goes down, we would have backup servers!
2. Horizontal Scaling
Horizontal Scaling is to increase the number of servers that we have. So basically, to handle more users, we increase the number of servers running in the backend and redirect the user request to different servers, so that the load gets balanced evenly!
In real-life systems, Scaling refers to a hybrid of these types i.e Horizontal + Vertical Scaling!
This brings us to the next topic: Load Balancing
3. Load Balancing
A load balancer is an Intermediate server that accepts all the user requests and redirect them to appropraite server based on certain algorithms, like Round Robin, Consistent Hashing etc. Load Balancer is sometimes also referred to as a Reverse-Proxy Server.
Sometimes, these multiple severs that we deploy are present in different geographical locations around the world, which brings us to the next topic: Content Delivery Networks or CDN's.
4. Content Delivery Networks(CDNs)
We have users all around the world, but generally, our backend(ORIGIN) server is located at a specific geographical location. So every time, when a user makes a request, this request will be slow if the user is located very far away from the server and fast vice-versa. So to tackle this problem, a copy of our server is created and distributed at different data centers all around the globe.
This reduces the response time and creates a network all around the globe. This is called Content-Delivery-Network. or CDN. CDNs are generally used to cache the static data such as images, videos, text etc. so that a group of users located at same geographical location, do not face any down time!
NOTE: These days CDNs are also being used for dynamic content delivery as well!
5. Caching
Caching, essentially means to create copies of out data, so that it can be fetched faster in future. Making network request is every time is usually expensive, hence browsers create a cache of data loaded in the Hard Disk. Reading HDDs can be expensive, hence out computer store a copy of that data in RAM. Reading RAM can even be expensive sometimes, so our Operating System, store this a subset of this data in L1, L2 and L3 CPU Cache.
But how do computer identify each other in a network?
6. IP Address
IP Address or Internet Protocol Address is the unique ID that is assigned to every computer on a network, so that computers can communicate with each other in the newtork.
7. TCP / IP Suite
TCP/IP suite is the set of protocols/rules that decides that how the data should be shared over the internet or over the network. As per TCP(Transmission Control Protocol), the data is shared in the form of numbered packets over the network. When these data packets reach the final destination, they get reassembled as per the numbers and the receiver is able to see the actual data shared!
NOTE: TCP/IP suite also contains UDP(User Datagram Protocol). TCP/IP is a complete model of how data is transferred from one computer to another. There's another famous model that co-exist with TCP/IP, which is the OSI(Open System Interconnection) model.
Ever wondered what happens when we write "google.com" on the browser!! How does the computer know which IP Address to communicate to when we write google.com !! This is where DNS comes into play.
8. DNS (Domain Name System)
DNS or Domain Name System is a large decentralized service that maps these Domain Names to its respective IP addresses. So, whenever we make a request to "google.com", our browser makes a DNS Query to get the IP Address of that domain name and then start communication with that IP! When browser makes a DNS query, it store that particular IP in its cache so that it doesn't have to make the DNS query again!
9. HTTP
TCP is a low level protocol in which we have to worry about all individual data packets being transferred, hence an Application Layer Protocol was devised which is HTTP(HyperText Transfer Protocol). HTTP is based on Client-Server architecture. Client Initiates a HTTP Request which contains two parts (Request Header + Request Body).
Request Header contains the data of where this request is going like the Server's IP, Content-Type, Authentication etc. Request Body contains the data/content which has to be given to the Server.
There are Multiple API Paradigms that can be followed with HTTP. Following are some of those:-
10. REST
REST or Representational State Transfer is a type of API that follows speicific guidelines for request and responses. In REST, request type is defined i.e. whether a request is GET, POST, PUT, PATCH or DELETE. Similarly, when server return a response, if returns the data with some a request code like 200 (Successful request), 404 (Page not found), 500 (Internal Sever Error) etc.
11. GraphQL
GraphQL or Graph Query Language is a type of API paradigm that was introduced by facebook in 2015 in which we pass a query to the server to get the multiple data fields that are required in a single request unlike HTTP, in which we have to make multiple request to get multiple data.Visit Official GraphQL website for more details.
12. gRPC
gRPC or general-purpose Remote Procedure Call is an API framework that was introduced by google to help microservices communicate faster. In general gRPSs are 7 time faster than REST APIs. This is because REST APIs uses JSON payloads where gRPSs use Protocol Buffers which are binary encoded data. Visit Official gRPC website for more details.
13. WebSockets
Websockets is the way to establish a bi-directional communication betwwen client and server. With HTTP, at one time, either client can request or server can respond to a previous request but with websockets asynchronously both request and response can travel at same time in a connection.
Now what is the need of websocket? Imagine a chat app as shown in the below figure!
Say, if client1 sends a message to client2, but how would client2 know that? If the connection is HTTP, client2 will have to poll the server in every 1 or 2 seconds to check whether it has a new message! This polling is what websockets cut out. As soon as client1 sends a message, server redirect that message to client2 in the websocket connection which in a way also make the system real-time!
Now, to store data, we need a database, the most popular being SQL Databases.
NOTE:- Database storage is more efficient that normal text files is because they store data efficiently using B-Trees and hence data push and retrieval also way more faster.
14. SQL
SQL or Structured Query Language is a relational database that stores data in the form of rows and columns. Ex:- MySQL, Postgres SQL etc. SQL Database are ACID compliant which stands for Atomicity, Consistency, Isolation, Durability.
NOTE: Consistency in SQL databases, makes them harder to scale since many foreign key constraints exits between tables in SQL databases.
15. ACID
16. NoSQL
NoSQL databases remove the consistency constraint from the SQL databAses, hence there is no specific schema that we have to define while storing data in NoSQL database unlike SQL which requires exact schema in which data has to be stored. Popular NoSQL databases are DynamoDB (key-value stores), MongoDB (document-collection), neo4j(graph based) etc.
17. Sharding
Since, NoSQL databases remove the consistency and foreign key constraints, the data can be broken down into multiple smaller databases and scaled horizontally for faster data write/retrieval. There are multiple ways in which a database can be sharded into smaller databases. One of the common techniques is to select a specific field from the data and make it a shard key, and use any hash function on this key to divide the database, as shown in the figure below!
NOTE: Database Sharding can at time become very complicated and hence a simpler option is to replicate the database to scale it globally.
18. Replication
In database replication, multiple replica of the database is created and distributed around the world for better scaling. There are two models that are followed in database replication:-
- Leader-Follower :- Here, Only Leader database, has both read/write permission whereas follower has only read permission.
- Leader-Leader :- All the entities of databases that are distributed has read/write permissions but it is very difficult to maintain consistency in this model.
NOTE: It is very difficult to keep all the replica of the databases in sync, so this is where CAP theorem comes into play!
19. CAP Theorem
CAP Theorem was incomplete and a more extended PACELC Theorem was created!
PACELC Theorem - GeeksforGeeks
20. Messages Queues
If our Server/Consumer receives more requests than it can process, the incoming requests are queued into a LinkedList and then processed by the server one by one. This is called Message Queue.
Top comments (0)