Snowflake-SI Algorithm

Snowflake-SI algorithm is a variety of Twitter's snowflake algorithm. The standard snowflake uses a 64-bits integer as output. However, the ECMAScript does not provides the supports of 64-bits integer. It's language standard shows that only integers between Number.MIN_SAFE_INTEGER and Number.MAX_SAFE_INTEGER(including both points), are accurate, which means only 53-bits integer can be used in ECMAScript.

The Snowflake-SI algorithm uses 53-bits integer as the output of UUID generation, its bits-mapping shows like following:

H	52 ~ 13	12 ~ 8	7 ~ 0	L
	1111111111111111111111111111111111111111	11111	11111111
	MAB	MID	UIN

As the graph, the low 8-bits is the UIN(UUID index number), which is the low 8-bits of an incremental-sequence. Thus, even if there is 256 generation in one millisecond, none of the generated UUID will be duplicated.

Theoretically, a single Node.js process can not handle 256 requests with logical IO operation. So that the capacity 256 UUIDs per millisecond can meet most of the cases. If 256 is not enough, please read the next section to adjust the algorithm.

The bits 8 ~ 12 are the MID (Machine ID), for distributed services. So that it will not generate duplicated UUID between multi-instances. Snowflake-SI use 5-bits for MID, so that there could be 32 instances at most.

The high 40-bits is the MAB (Milliseconds after epoch). Before using the Snowflake-SI algorithm, a epoch must be determined, which is the base to start time of using the UUID generation. The MAB is the number of milliseconds after the epoch. To avoid generating duplicated UUID, the MAB must be an incremental sequence, and then the epoch is unchangeable after it's determined.

According to the max time that UNIX timestamp can represent is 2038-01-19T03:14:07Z, 0x7FFFFFFF, and 2147483647000 in milliseconds. It's a 41-bits integer, but the MAB must be a 40-bits integer. That's why it uses MAB instead of timestamp. And, for this, the epoch must meet a condition that 2147483647000 - BaseClock < Math.pow(2, 40). It means BaseClock > 1047972019224, while 1047972019224 stand for 2003-03-18T07:20:19.224Z.

Algorithm Adjustment

As above said, the high 40-bits is fixed, but the rest 13-bits is adjustable. That means if distributed deployment is not necessary, the MID is avoidable, so the whole 13-bits can be use as the UIN, while its capacity is up to 8192

And actually, the MAB could be adjusted, too. If the bit-width of MAB is set to 41-bits, so the epoch could be more earlier, but at the same time, the capacity of UUIDs per millisecond and machine identifiers will be reduced.