Why do NoSql databases scale better than relational databases? How should I choose between them?

user_7917681 0
Viewed 825

By nosql databases I mean something like mongodb or dynamodb I've been trying to find why NoSql dbs usually are usually better at horizontal scaling than relational dbs, and how to choose between them

I have looked into many videos and posts that tell us the "SQL vs NoSQL". Most of them end up talking about "Normalization vs Denormalization".

Here are some questions I am still confused about.

1. Many people said that relational dbs have to follow ACID so they are bad at horizontal scaling. But ACID is about transaction, we can always choose not to use any transaction, right? I know not many people do this, but if we denormalized tables enough, would it be like NoSQL dbs where we almost don't use any transaction?. And many NoSql dbs now have transactions too.

2. I know denormalization is probably good for horizontal scaling, because if data are spreaded across many nodes(machines), it'll be hard to do table joining(or transaction). But like transaction, we can choose not to use any table join.

The only thing I can think of is NoSQL are schema-free, it is easier to add new fields(columns) than RDB.

What I am trying to ask are
why is a "Denormalized NoSQL db" better than a "Denormalized relational db" ?
why is a "Normalized NoSQL db" worse than a "Normalized relational db" ?
what's the real thing that prevents relational database from denormalization?

I've read this post
https://softwareengineering.stackexchange.com/questions/194340/why-are-nosql-databases-more-scalable-than-sql
It says
""The SQL API lacks a mechanism to describe queries where ACID's requirements are relaxed. This is why the BASE databases are all NoSQL.""
Could anyone give me an example of this?

Sorry for not being specific

By NoSQL databases I mean something like mongodb

mongodb nosql mysql sql database
2 Answers

A blog like https://neo4j.com/blog/acid-vs-base-consistency-models-explained/ explains BASE this way:

Basic Availability

  • The database appears to work most of the time.

Soft-state

  • Stores don’t have to be write-consistent, nor do different replicas have to be mutually consistent all the time.

Eventual consistency

  • Stores exhibit consistency at some later point (e.g., lazily at read time).

This level of equivocation doesn't sound very reliable, does it? They trade off availability and consistency to gain performance and scalability.

This is fine if you're running a service that is tolerant of mismatched data or stale data, or which is okay with some minor amount of data loss once in a while. If those issues are an uncommon occurrence, but you get superior performance nearly all the time, it's very attractive. And more importantly, it demos well.

But if you have to run a service with strict requirements for data integrity, it's no good. If losing even one record of data gets you in trouble with auditors, or if you can't reliably read data you just committed a moment before because that commit takes time to propagate to all nodes of your cluster, it could be a deal-breaker.

So which data store to choose depends on the requirements of your app. Only you can judge if the relaxed availability and consistency of a BASE data store is sufficient for the needs of your app.

Bill_Karwin_208600

NoSQL is a term that covers lots of types of storage/query engines e.g. document stores, Graph Databases, etc. - basically anything that looks something like a database but doesn’t use the standard tables/rows/columns structure that a SQL database does.

NoSQL databases were developed to support use cases that relational databases don’t handle well - so while you might be able to use either a SQL or a NoSQL database in any given scenario, the choice between the 2 is normally a no-brainer; they would very rarely both be viable options.

Just to clarify, your questions about types of DB being better or worse are meaningless without context. Without knowing precisely what your requirements are, it’s impossible to say whether a NoSQL DB is better or worse than a SQL one - and that’s before you start looking at specific products in each category.

Also, that post you reference is about 8 years old and much of the information is out of date - as one of the contributors acknowledges in an update made in 2019

user_136583990
Related