Basic System design overview

Tung Thanh — Sat, 08 Jul 2023 07:39:24 +0000

Every one already know that system, so what happen if we want to adopt this system to serve million requests.

Keywords: Rate Limit, DNS, Proxy, Reverse proxy, Load Balancing, CDN, Blob storage, Vertical Scaling, horizontal scaling.

What should I do when a user spam requests?

At that time, we need to have a rate limiting. We can restrict the number of requests that a client can make to a server within a period of time.
AWS provides several services that can be used for rate limiting, depending on the specific needs of your application: API Gateway, CloudFront, AWS WAF (firewall), AWS Lambda.

where do static files(html, css) store?

If we store it in database so each client's request will need to go to database that make the server cannot handle.
That's why we need to have another server such as BLOG storage (S3, Azure Blog,...) and then use a CDN service (Cloudfare, AWS Cloudfront, GCP Cloud) to deliver static files for load reduction on the main server.

How do we scale the main server?

Vertical scaling - scale up

Increase the power of server: add more rams, upgrade CPU,...

Horizontal scaling - scale out

Increase the capacity of a system by adding more machines (nodes)

When scaling out, how does the client know which servers they should communicate with?

The Load balancer is responsible for distributing incoming requests from clients to the available servers in a way that ensure optimal resource utilization and performance.
Or for more simpler, we just need to have a reversed proxy to handle.
(Actually reversed proxy can be used as a LB).
LB and reverse proxy are 2 distinct technologies that can be used together and to improve performance and scalability of web app. They have several different purposes and different capacities.

After that, we have a system like below.

Bottlenecks at Database

When the number of requests from servers is high, the system may still experience bottlenecks at the database levels.

1. Data not change frequently

--> We need to have a cache database
Instead of getting data from database, server will read from cache first
-> improve read speed because cache data is stored in memory.
There're 2 types of caching:

In-Memory Cache:
- Stores in RAM of a single server or node in a network.
- Problem is when RAM capacity has been exceeded so we need to use cache eviction algorithms: LRU, FIFO, LFU.
Distribution Cache
- Cache is share across multiple servers or nodes in networks.
- But if the cache server is shutdown, we would loss all cache data, so to make the system high availability (HA), the cache should be replicated to multiple nodes (like master-slave strategy,...).

We even combine 2 kind of caching and in multiple levels:
Flow: get Cache from In-memory -> cache miss -> Get from Distribution - > Miss -> Get from DB.

Indexing and Key in Database

Tung Thanh — Sun, 25 Jun 2023 15:26:53 +0000

Note

Indexing → care about the distinct value of the column
- more duplicated value → low performance
Index affects to IS_NULL operator
- when this column is not indexed → needs a table full scan to find null values.

Why we need indexes for Database tables

Benefits

Speed up searching.
Indexing helps in faster sorting and grouping of records.

Drawbacks

Additional disk space
- The clustered index doesn’t take any extra space as it stores the physical order of the table records in the DB.
- Non-Clustered Index needs extra disk space.
Slower data modification
- update record in the clustered index

Overview

The index is nothing but a data structure that store the values for a specific column in a table (an index is created on a column table).
Improve the speed of data retrieval operations.
With DML operations, indices are updated, so write operations are quite costly with indexes.
- The more indices you have, the greater the cost.
- Indexes are used to make READ operations faster.
- So if you have a system that is written-heavy but not read-heavy, think hard about whether you need an index or not.
Cardinality is IMPORTANT
- Cardinality means the number of distinct values in a column.
- If you create an index in a column that has low cardinality, that’s not going to be beneficial since the index should reduce search space. Low cardinality does not significantly reduce search space.

Clustered and Non-Clustered index

A clustered index is a table where the data for the rows are stored

Each table has only one clustered-index - that stores row data

When we define PK → InnoDB use it as the clustered index
If we don’t define a PK → It will use the first UNIQUE index in this table
If a table has no PK or suitable UNIQUE index → It will generate a hidden clustered index: GEN_CLUST_INDEX.

Each record in a secondary index contains the PK columns for the row as well as the columns specified for the secondary index.
All InnoDB indexes are B-trees where the index records are stored in the leaf pages of the tree.

The Default size of an index page is 16KB MySQL::InnoDB Page Size The MEMORY storage engine (known as HEAP) supports both HASH and BTREE index → creates special purpose tables with contents that are stored in memory. In this engine, there
HASH for equality operator (only available on MEMORY engine)
BTREE for range operator (both in MEMORY and InnoDB)

Forem: Tung Thanh