Use cases - NoSQL

• If the requirements are needed, use NoSQL rather than SQL database
  • low latency
  • unstructured data
  • large size data
  • only serialize or de-serialize data (JSON, YAML, XML)

Scale out vs Scale up

• Scale up: Same number of machine, update the unit’s of a machine’s tier. The performance of system is updated vertically.
  • eg. update CPU instance or memory of the machine
  • pros
    • simple
  • cons
    • the limit, we can not update the parts of a machine indefinitely
    • there is no failover / redundancy strategy
• Scale out: Add more server, update the number of server (instance). The performance of system is updated horizontally.

Cache

External link

‣

• When we should use cache?
  • Frequently read, rarely updated data
• Which data should be in cache?
  • Basically, the cache layer is volatile
  • So set data which is not important
• Cache management - how to evict them?
  • LRU (least recently used)
  • LFU (least frequently used)
  • FIFO (first in first out)
• Consistency
  • If cache-update calculation and storage-update calculation are not wired up, the consistency should be broken.
  • REF. Scaling Memcache at Facebook - Facebook
• SPOF, Single Point of Failure
  • If there is only one cache server, the server should be a SPOF, single point of failure.
    • Definition of SPOF: A single point of failure (SPOF) is a part of a system that, if it fails, will stop the entire system from working - Wikipedia, SPOF

• Key points of system design interview.