Fundamentals of System Design.
Monlith and Microservices.
Monolith doesnot need to be single in terms of number of machine we are running it on.
There can be multiple machines in monolith and the client can connect to any machines.
We can horizontally scale out in Monolith. Client directly talks with the monolith server.
It is easy for small team and less work.
Less duplication of code.
It is faster. No need to make calls ovr the network. Procedure call is faster.
All in the same machine so faster.
- Deploment is complex. Any change the entire system needs to deploy.
- Single point of failure. Multiple monolith server can work for horizontal scaling. So if anything goes wrong the server will be down.
Microservice is a single business unit. All data which are relevant to the service are in one service. We can separate services in pieces.
Like there is 3 Microservice in the system and they talk to the dedicated db.
Client connected to the Gateway and gateway connect to the Microservices.
- Easy to deploy. any new member will find easy to work in specific section.
- Architecture design is imp. If one server S1 is always talking to the other service S2 then there should no rtpc call and it should be only function call.
Cloud computing where the cloud service provider provides some computers for the computational purpose that act as a server.
When we get more request to handle the request we can Buy bigger machines means Vertical Scaling and Buy more machines means Horizontal Scaling.
| Horizontal Scaling | Vertical Scaling |
|---|---|
| LoadBalancing required. More machines which machine will handle which request should be determined by Load Balancing. | Not required. |
| Can handle request when one server fails. | Single point of failure. |
| More network calls(RPC Remote Procedure Calls.) It is slow. | Inter process communication. It is fast. |
| Data needs to travel to all the server and it is loose couple. To make the transaction atomic we need to stop all the other server works and it is not possible. Data inconsistent. | Consistent. |
| Scales well. | Hardware Limit. |
LoadBalancing and Consistent Hashing.
Lot of request is coming say the request id is from 0 to M-1.
The request id is send to the server. We take the request id r1 and then we hash it.
There are some hashing algorithm.
Say h(r1) = m1 then to redirect we m1%n n is the number of server we have. Then we send the request to the respective server.
Say we have 4 server and after h(10) - 3 say the hashing is making the value 3.
Then 3%4 = 3 So teh request will go to the 3rd server.
Hash function is random. So all server will have uniform load.
If there is X request each of the server will have x/n load and the load factor is 1/n.
The request is generally not random it depends on the user id or anything.
Example. Say h(39) = 15 and 15%4 = 1 Now the request will go to the 1st server.
Now the problem will arise when we want to add another server. Now that the request id is same when we change the n value the server number will change.
Generally for specific request fixed server takes care of it and the data is present in the cache of the server. Now everything is changed.
Now for this example h(39) = 15 and 15%5 = 0. Server count is 5.
We can see that in Consistent Hashing.
Pie diagram.