Medo.Uuid7 1.6.0

Medo.Uuid7

The UUID7 library is an implementation of the UUID version 7 variant as defined in the new UUID specification, which introduces a time-ordered value field derived from the timestamp source. It offers improved entropy characteristics compared to versions 1 or 6 of the UUID standard. The inherent monotonicity of UUID version 7 makes it an excellent choice for utilization as a binary database key.

Features:

• Time-ordered value field: UUID7 utilizes the widely implemented Unix Epoch timestamp source to generate a time-ordered value field. This enables easy sorting and indexing of resources based on their creation time.
• Enhanced entropy characteristics: UUID7 provides improved entropy characteristics over UUID versions 1 or 6. The inclusion of the timestamp ensures a high level of uniqueness, minimizing the chances of collisions across different systems or instances.
• Multiple string representations: In addition to the standard UUID string formatting, library also offers ID22 and ID25 string conversions.
• Wide compatibility: Support for .NET Standard 2.0 makes this library compatible with .NET Framework 4.6.1 or higher.
• High performance: Speed comparable to the optimized built-in GUID generator in both single-threaded and multi-threaded scenarios under Windows and Linux.
• Hardware acceleration: Vector128 support for Equals method.
• Microsoft SQL Server support: Separate methods for "LE" Guid creation.

You can find packaged library at NuGet. Also available is an Entity Framework Core version.

Usage

To generate a new database-friendly UUID v7, simply call NewUuid7 method:

using System;
using Medo;

var uuid = Uuid7.NewUuid7();  // or 'new Uuid7()' or 'new Guid()'
Console.WriteLine($"UUID : {uuid}");

Alternatively, if a fully random UUID v4 is desired, call NewUuid4 method:

using System;
using Medo;

var uuid = Uuid7.NewUuid4();
Console.WriteLine($"UUID : {uuid}");

Configuration

Disable RNG Buffering

Buffering of random numbers significantly increases performance at the cost of less frequent but bigger requests toward random number generator. If buffering is not desired (e.g. only a small count of UUIDs is needed), you can disable it using UUID7_NO_RANDOM_BUFFER preprocessor constant.

<PropertyGroup>
    <DefineConstants>UUID7_NO_RANDOM_BUFFER</DefineConstants>
</PropertyGroup>

Note that this will decrease performance significantly.

UUID Format

The format of UUIDv7 is as specified below.

 0                   1                   2                   3
 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                           unix_ts_ms                          |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|          unix_ts_ms           |  ver  |       rand_a          |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|var|                        rand_b                             |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                            rand_b                             |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

unix_tx_ms: 48 bit big-endian unsigned number of Unix epoch timestamp.

ver: 4 bit UUIDv7 version. Always 0111.

rand_a: 12 bits of pseudo-random data.

var: 2 bit variant. Always 10.

rand_b: Additional 62 bits of pseudo-random data.

Implementation

As monotonicity is important for UUID version 7 generation, this implementation implements most of monotonic random counter recommendations.

Implementation uses randomly seeded 26 bit monotonic counter (25 random bits + 1 rollover guard bit) with a 4-bit increment.

Counter uses 12-bits from rand_a field and it "steals" 14 bits from rand_b field. Counter will have its 25 bits fully randomized each millisecond tick. Within the same millisecond tick, counter will be randomly increased using 4 bit increment.

In the case of multithreaded use, the counter seed is different for each thread.

In the worst case, this implementation guarantees at least 2^21 monotonically increasing UUIDs per millisecond. Up to 2^23 monotonically increasing UUID values per millisecond can be expected on average. Monotonic increase for each generated value is guaranteed on per thread basis.

The last 48 bits are filled with random data that is different for each generated UUID.

As each UUID uses 48 random bits in addition to 25 random bits from the seeded counter, this means we have at least 73 bits of entropy (without taking 48-bit timestamp into account).

With those implementation details in mind, the final layout is defined as below.

 0                   1                   2                   3
 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                           unix_ts_ms                          |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|          unix_ts_ms           |  ver  |        counter        |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|var|          counter          |            random             |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                            random                             |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

unix_tx_ms: 48 bit big-endian unsigned number of Unix epoch timestamp.

ver: 4 bit UUIDv7 version. Always 0111.

var: 2 bit variant. Always 10.

counter: 26 bit big-endian unsigned counter.

random: 48 bits of random data.

Textual Representation

While this UUID should be handled and stored in its binary 128 bit form, it's often useful to provide a textual representation.

UUID Format

This is a standard hexadecimal representation of UUID with dashes separating various components. Please note that this component separation doesn't necessarily correlate with any internal fields.

Example:

0185aee1-4413-7023-9109-bde493efe31d

Id25

Alternative string representation is Id25 (Base-35), courtesy of stevesimmons. While I have seen similar encodings used before, his implementation is the first one I saw being used on UUIDs. Since it uses only numbers and lowercase characters, it actually retains lexicographical sorting property the default UUID text format has.

UUID will always fit in 25 characters.

Example:

0672016s27hx3fjxmn5ic1hzq

Id22

If more compact string representation is needed, one can use Id22 (Base-58) encoding. This is the same encoding Bitcoin uses for its keys.

UUID will always fit in 22 characters.

Example:

1BuKkq6yWzmN2fCaHBjCRr