Skip to content
resizable hashing strategy for large-scale storage
JavaScript
Branch: master
Clone or download
Latest commit d4bfdd6 Oct 6, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
example using builtin bigints Oct 4, 2019
test use bigint-rational Oct 6, 2019
LICENSE-APACHE documentation and license Oct 4, 2019
LICENSE-PARITY documentation and license Oct 4, 2019
contributing.md documentation and license Oct 4, 2019
index.js reuse some rational allocations Oct 6, 2019
package.json 2.0.2 Oct 6, 2019
readme.md serialize and parse Oct 4, 2019

readme.md

random-slicing

resizable hashing strategy for large-scale storage

implements the algorithm from the random slicing paper, a better alternative to the hash ring technique used in riak

The random slicing algorithm is designed to maintain a resizable address space across storage nodes while minimizing the amount of data that needs to be moved during a resize operation on the address space.

The address space exists on a real number line from 0 to 1. Each bin is granted slices on this number line based on its size. These allotments may change over the course of the program, but the algorithm will minimize the change in slices for each bin while preserving the size ratios.

This implementation internally uses arbitrary-precision rationals for slicing calculations to eliminate rounding errors as the system evolves over time. Consult the hash example for how to convert these rationals into the hash space of your chosen hashing algorithm.

example

In this example, we initialize a previous allocation for nodes A (size 40) and B (size 120). In practice you might get this previous allocation from persistent storage or the network.

Then, we shrink A from 40 to 32 and add a new node C with size 80.

Finally we display the integer ratios for each slicing interval.

var RS = require('random-slicing')
var rs = new RS
rs.set({ A: 40, B: 120 })
rs.set({ A: 32, C: 80 })

Object.entries(rs.getBins()).forEach(function ([key,bin]) {
  console.log(key, bin.size, bin.slices.map(showSlice).join(', '))
})

function showSlice ([start,end]) {
  return `${start[0]}/${start[1]}..${end[0]}/${end[1]}`
}

which prints:

A 32 0/1..640/4640
B 120 40/160..3560/4640
C 80 640/4640..40/160, 3560/4640..160/160

api

var RS = require('random-slicing')

var rs = new RS(bins)

Initialize a new random slicing with an optional allocation of bins.

bins should be of the format returned by rs.getBins() documented below.

rs.set(updates)

Set the new sizes with an object updates mapping keys to their new sizes. Keys not present in updates will keep the same size.

To delete a bin, set its size to 0.

var bins = rs.getBins()

Return the collection of allocated bins, an object that maps bin names to bins, where each bin has:

  • bin.size - presently allocated size
  • bin.slices - array of intervals

Each interval is an array 2-tuple [start,end] and start and end are each array 2-tuples of the form [numerator,denominator] where numerator and denominator are both built-in bigints.

For example, a bin might look like:

{
  size: 10,
  slices: [[[0n,1n],[1n,7n]],[[87n,364n],[1n,4n]]]
}

which contains slices from 0 to 1/7 and from 87/364 to 1/4.

var str = rs.serialize()

Serialize an rs instance to a string.

var rs = RS.parse(str)

Create a new rs instance from a previously serialized string str.

license

license zero parity and apache 2.0 (contributions)

You can’t perform that action at this time.