Design Requirements

• Functional requirements: These represent the features a user of the designed system will be able to use. For example, the system will allow a user to search for content using the search bar.
• Non-functional requirements (NFRs): The non-functional requirements are criteria based on which the user of a system will consider the system usable. NFR may include requirements like high availability, low latency, scalability, and so on.
• Requirements Definitions

Performance vs scalability

A service is scalable when remain performant in proportion to the resources added

• If you have a performance problem, your system is slow for single user
• If you have a scalability problem, you system is fast for a single user but slow under heavy load

Latency vs throughput

• Latency is time take from perform some action and receiving the result
• Throughput is the number of such action or result per unit of time
• Generally, you should aim for maximal throughput with acceptable latency.

Availability vs consistency

CAP Theorem

• Consistency - Every read receives the most recent write or an error
• Availability - Every read receives a response, without guarantee it contain the most recent version of the information
• Partition Tolerance - The system continues to operate despite arbitrary partitioning due to the network failures

CA

• Cannot exist in real-world applications due to the unavoidable network failure

CP

• Waiting for a response from a partitioned node might result in a timeout error.
  • Good choice if requires ATOMIC read and writes
  • Banking system

AP

• Responses return the most readily available version of the data available on any node, which might not be the latest.
  • AP is a good choice if the business needs to allow for eventual consistency or when the system needs to continue working despite external errors.

Consistency Pattern

Weak consistency

• After a write, read may or may not see it.
  • Memcached
  • UDP

Eventual consistency

• After a write, reads will eventually see it (typically in milliseconds). Data is replicated asynchronously.
  • DNS and email
  • Dynamo and Cassandra

Strong consistency

• After a write, reads will see it. Data is replicated synchronously.

Availability patterns

Fail-over

• Active-passive
  • Heartbeats are sent between the active and passive server. If the heartbeat interrupted, the passive server takes the IP address and resume services
  • The length of downtime is determined by whether the passive service is already running, or needs to start up
• Active-active
  • Both server are managing traffic
  • DNS should know both IPs

Disadvantages

• Fail-over add more hardware and additional complexity
• Potential loss of data if the active system fail, before any newly written data can be replicated to the passive

Replication

Availability in numbers

• Three 9s
  • 1m 26.4s downtime per day
• Four 9s
  • 8.6s downtime per day