Distributed Unique ID Generator

clarifying questions.
- ids must be unique and sortable.
- ids are only numerical values.
- ids should fit into 64 bit.
- system should generate 10,000 IDs per second.
possibilities.
- scale id values by k where k is the number of servers.
  - disadvantages.
    - this does not scale well when we want to add or remove servers.
- generate unidue uuid’s of 128bit length, very unlikely to have a collison.
  - disadvantages.
    - these ids cannot be sorted by time of creation.
    - they contain values that are not numeric.
- have a single server that generates ids for each of the distributed servers.
  - disadvantages.
    - single point of failure.
- - we can create custom ids by mixing a bunch of various ids including the timestamp, datacenter id, machine id, and a sequence number that is incremented with each id and reset every millisecond.
deep dive - twitter snowflake approach.
- by using a timestamp at the beginning, our ids can be sorted by time.
- by using data center and machine ids, we avoid collisions in a distributed system.
- we use the last 12 bits for generating a sequence number, to help distinguish ids that are generated on the same timestamp.
- advantages.
  - high availability because no servers depend on any servers to run, so single point of faliure.
  - ids can be sorted by time created.

🗿