Chapter 9: System Design and Scalability

Networking Metrics

-Latency -Bandwidth -Throughput

Algorithms that scale: step-by step

1) Ask questions 2) Make believe by making assumptions 3) Figure out how to do this reasonable 4) Solve your potential issues by either removing it, or mitigating the problem

Design Outline for systems

1) Scope the problem 2) Make reasonable assumptions 3) Draw major components 4) Identify key issues 5) Redefine key issues

Load Balancer

Distributes information aroud the nodes/server so that one server doesn't crash and take down the whole system. Most likely has systems containing cloned data(redundancy)

Considerations

Failures Availability and reliability Read-heavy vs Write-Heavy Security

Latency

How long it takes from data to go from one end to another. The delay between the sender and reciever.

Bandwidth

Maximum amount of data that can be transferred in unit of time

Vertical Partitioning

Partitioning by feature. Example: One partition has messages, another has profiles, etc

Cache

Small database that is close to the CPU so rapid lookup is possible. If it misses, it will request data from the server or harddrive.

Throughput

The actual amount of data that is transferred in a given time

Denormalization

To add redundant information into a database to speed up reads. Joins in a large database get extremely long, so you might want to denormalize it and add common information(like table name) to the table.

Horizontal Scaling

To increase the number of nodes

Vertical Scaling

To increase the resources on one node

Key-base(or hash-based) partition

Use some part of the data to partition. Limits the number of serves

Mapreduce

Uses a Map step and a reduce step. The map takes in some data and emites a <key, value> pair The Reduce portion takes that key and associated value, and "reduces" them in some way, emitting a new key and value.