System Design Interview ext: Article Link

Database

ACID

• Atomicity - each transaction is all or nothing
• Consistency - the database stays valid between transactions
• Isolation - concurrent transactions has the same results as serial
• Durability - once a transaction completes, it remins complete

Replication

Master-Slave

• The master serves reads, the slaves only serve writes

Master-Master

Both serve reads and writes and coordinate on writes

Disadvantages:

• Requires a load balancer
• Either violates ACID (loosely consistent) or requires slow synchronization logic

Federation (Functional Partitioning)

• Splitting up the database by function - ex. forums, users, products

Disadvantages

• Complex joins
• Complex queries over multiple dbs

Sharding

• Split the data over multiple dbs

Disadvantages

• Complex queries
• Complex joins

Denormalization

• Improve reads, hinder writes

Redundant copies of data are written in multiple tables to avoid expensive writes

This strategy works in conjunction with federation and sharding

SQL Tuning

• Using CHAR instead of VARCHAR for fixed-length fields
• INT for large numbers up to 2³²
• DECIMAL for currency
• Set the NOT NULL constraint wherever possible
• Use good indices
• Denormalize data that will be frequently joined
• Partition hot spots

Indices

• Represented as self-balancing B-trees

Caching

• Whenever your application tries to read data, it should first look through the cache

Two Approaches:

1. Cache the result of database queries

2. (Recommended) Cache objects

   • Store the complete class in the db, or
   • Store the arrays, etc. in the db

Cache:

• User sessions
• Blog articles
• user-friend relationships

Asynchronism

• Have worker nodes that constantly check from a message queue

• Once it's done, they send a completion message

  Anything time-consuming, do it async

Latency vs. Throughput

• Latency - the time to perform an action
• Throughput - the number actions / time. Ex. 120 cars per day

Availability vs. Consistency

A system can only support two of the following:

• Consistency - Every read receives the most recent write
• Availability - Every request receives a response
• Partition Tolerance

Consistency Patterns

• Weak Consistency - Reads may or may not see it
• Eventual consistency - Reads will eventually see the write
• Strong Consistency - Reads will see it

Availability Patterns

Fail-over - Switching reliably between backup systems

• Active-passive - Heartbeats are sent between systems. If it's interrupted, the passive server takes over
  • Aka master-slave failover
• Active-Active - Both servers manage traffic

Delivery of Information

DNS - Translates a URL to an IP address

• These can come under DDOS attack

• More laggy than CDN

  CDN - A global network of proxy servers. These server content to users closer to their locations

• Rewrite your URL to point to the CDN

  • Push CDNs - Receive new content when changes occur on the server
  • Pull CDNs - Grab new content from the server when the user requests it
    • This is a slower request

Load Balancer

• Distributes user requests among clusters of servers
• Servers contain no session info, this should be in Redis or db

Reverse Proxy

• Centralizes client requests. This sits between the client requests and the web server

Lets you have more, limit connection per client, blacklist IPs

Nginx is an example of [a] reverse proxy server

• Make your servers portable
• Compress server responses

Load Balancer vs. Reverse Proxy

• Load balancer is good for horizontal scaling with multiple servers
• Reverse proxies are good with one server

Platform Layer

• Sits between the web servers and the Database

NoSQL

Cache

Asynchronism

RPC

A client causes a procedure to execute on a remote server. RPC abstracts the method call to look exactly like what it would be on the client

RPC sends data using a binary codec implementation (protobuf, thrift, avro)

REST