System Designing

Hey Guys, 

This is such an interesting topic and you might have come across it in the interview process, so I've been going through various tutorials to complete the system designing concepts. This will check the candidate's ability to build a complex and large-scale system, and this system designing round in interview is not much focused on coding. 

System designing is nothing to do only with the database or backend. It is all related to high-level designing. It uses concepts from computer science such as networks, distributed systems, and parallel computing to build large-scale systems like Facebook, Google, Amazon, and Tiktok. 

Client-Server Model

Server & Client explanation is given in this post. 

Data or code will not be always stored in the file or the database of a computer/server.  

We can send this via the internet using protocols. The data or code can be hosted on the cloud too. 

APIs are used to send and receive requests and responses. When one computer starts sharing the data it holds with others, the client-server concept has begun. 

Eg: If an application needs to reach millions of users, we should need brilliant engineering on the server-side. So the server should never fail. Even if the hardware fails, we should have multiple copies of the same server in the system. Changing the number of copies should be easy, we can add or remove the servers. Each server in the system should be the same while maintaining multiple copies. If not the same, the user might get contradicting information while making a request to a different server. 

Cloud is a set of computers which was bought for a price. If we pay a cloud service, for example AWS, AWS will provide computation power. AWS is a solution provider where the configuration, settings, reliability can be taken care of to a large extent. 

Computation power is a desktop that runs my/our algorithm. If we remotely log in, we can use the services. 

System Design Concepts 

1. Caching

2. Load Balancing 

3. Partitioning

4. Proxies 

5. Messaging Queue or Message Passing

6. CAP Theorem 

7. Databases  (NoSQL Databases)

Caching

Why we need a cache? 

When the number of requests is more, there will be a lot of reads and write operations on the database which slows down the performance of the system. To handle these requests, a cache is used by reducing the need to access the underlying slowest storage layer.  

We used to save a lot of grocery essentials in our pantry which would save the time of cooking and a lot of time. Similarly, if we need to access a certain piece of data often, we need to store it in cache and retrieve it. Because accessing the data from the primary memory(RAM) is faster than accessing the data from secondary memory(disk).  

Load Balancing

The server, on which the code/software runs, will not be able to handle a lot of requests or load from billions of users. If it receives more requests than its capacity, the server might slow down. At low as possible, the load needs to be balanced evenly on the server. This is called load balancing. The concept of scalability is how to handle more requests, and load balancing is who will be distributing this high number of requests. In short, the load balancer is the traffic manager. 

Load balancer's job is to distribute the requests to many different servers to help with throughput, latency, performance, and scalability.  This will also ensure that server is not overloaded. Load balancing is explained here in detail.

Scalability 

Buying more machines to handle more requests or buying bigger machines to hanlde more requests is called scalability. 

Buying bigger machine - Vertical scaling 

Buying more machine - horizontal scaling 

Horizontal Scaling 

By buying more machines, the requests will be randomly distributed among all the machines. 

• It requires some load-balancing 

• Resilient - If one machine fails, the request may/not will be redirected to others. 

• All the communication is over the network. (Network calls - RPC), so it is slow. 

• Data inconsistency 

• It scales well when the number of users increases. 

Vertical Scaling 

All the requests will be sent to the bigger machine. 

• No load-balancing is required. 
• Single point of failure. 
• Inter-process communication, so this is fast. 
• One system where all the data resides, so this is consistent.  
• Hardware Limitation - We can't keep making the desktop bigger. 

Databases

We must also ensure that requests are responded to in a timely fashion. This requires optimization of networks and databases. 

CAP Theorem

Consistency majorly plays a role in distributed systems.  It is usually related to Data. It often included in CAP(Consistency, Availability, Partition Tolerance) theorem where C stands for consistency. 

<will be updated>