System design: How Facebook Scale its Social Graph Store? TAO

 Here is the article. 

Social graph data is stored in MySQL and cached in Memcached

3 problems:

1. list update operation in Memcached is inefficient. cannot append but update the whole list.
2. clients have to manage cache
3. Hard to offer read-after-write consistency

To solve those problems, we have 3 goals:

• online data graph service that is efficiency at scale
• optimize for read (its read-to-write ratio is 500:1)
  • low read latency
  • high read availability (eventual consistency)
• timeliness of writes (read-after-write)

Data Model 

• Objects (e.g. user, location, comment) with unique IDs
• Associations (e.g. tagged, like, author) between two IDs
• Both have key-value data as well as a time field

Solutions: TAO 

1. Efficiency at scale and reduce read latency

   • graph-specific caching
   • a standalone cache layer between the stateless service layer and the DB layer (aka Functional Decomposition)
   • subdivide data centers (aka Horizontal Data Partitioning)

2. Write timeliness

   • write-through cache
   • follower/leader cache to solve thundering herd problem
   • async replication

3. Read availability

   • Read Failover to alternate data sources