Unique identifiers for data and servers

[tshemsedinov]

out of date

GSID (Global Storage Identifier) is a number that can be represented in tree to map GSID to spatial-distributed servers. Any binary branch may fork or not at any bit to twice storage.

• When we have 1 server, everything stored there;
• For 2 servers, lower bit in GSID shows where stored data;
• For 3 servers, two lower bits shows storage place, for example: 00 for server S0, 01 for server S1, 10 for server S2 and 11 for server S2; or we may have different case: 00 for server S0, 01 for server S0, 10 for server S1 and 11 for server S2;
• For 4 servers, two bits encodes 4 different storage servers;
• When we want to increase number of servers, we should separate data at certain server based on next GSID bit and copy half of data to new server;
• Servers are mapped to geo-location (spatial sharding):
  • example for 4 servers:
    • 0 (0000) Europe and Africa,
    • 1 (0001) India and Australia,
    • 2 (0010) China, Japan, Korea and the neighbors
    • 3 (0011) the Americas,
  • then we fork into 8 servers:
    • 0 (0000) Europe,
    • 1 (0001) India,
    • 2 (0010) China,
    • 3 (0011) North America,
    • 4 (0100) Africa and the Middle East,
    • 5 (0101) Australia and Southeast Asia,
    • 6 (0110) Japan and the neighbors,
    • 7 (0111) South America.
  • and so on, but we can upgrade 4 servers to 6, e.g.:
    • 0 (0000) Europe and Africa,
    • 1 (0001) India,
    • 2 (0010) China,
    • 3 (0011) the Americas,
    • 5 (0101) Australia and Southeast Asia,
    • 6 (0110) Japan and the neighbors.
• So we need servers map because we do not know where servers forked and where not; such map will be stored at all servers and we will have GSID resolution service for finding a place of data physical storage.
• GSID is an integer number from 0 to ∞, with hexdecimal string representation like this 0xA83FB217;
• Servers are also have GSID, e.g. China is 0x02 from example above;
• Current hardware type for GSID is 64-bit integer 0x0000000000000000;
• Sometimes we can use binary GSID e.g. 0b10101000001111111011001000010111;

[tshemsedinov]

Closes in favor of metarhia/Metarhia#21