Mateo Sanabria Ardila

Conflict-Free Replicated Data Type (CRDT)

peer to peer COLLABORATIVE   TExt editing

Les't remember some CS foundation

The CAP THEOREM

(CAP) a distributed data store can provide only two of the following three guarantees:
  • Consistency: Every read receives the most recent write or an error

  • Availability: Every request receives a (non-error) response, without the guarantee that it contains the most recent write

  • Partition Tolerance: The system continues to operate despite an arbitrary number of messages being dropped (or delayed) by the network between nodes

Consistency

Availability

Partition  Tolerance

Achieving internet-scale distribution with low latency and high availability 
Strong Consistency
  • Requires coordination for executing operations.
  • Challenges arise during network partitions or failures.
Weak Consistency
  • Allows temporary divergence among replicas.
  • Requires merging concurrent updates into a common state.
Weak Consistency
WC
Strong Consistency
SC

Eventual Consistency
EC

Eventual Consistency

Replicas will eventually converge to the same final value if no further updates are submitted:
Eventual Delivery An update sent to any correct replica is guaranteed to be eventually received by all correct replicas.

Convergence Correct replicas, having received the same updates, will eventually reach an equivalent state.
Weak Consistency
WC
Strong Consistency
SC

Eventual Consistency
EC

Weak Consistency
WC
Strong Consistency
SC

Eventual Consistency
EC

?

  • Some EC systems execute updates, identify conflicts later, and roll back, leading to wastage.
  • Resolving conflicts often requires consensus among replicas.
  • A stronger condition is needed to improve efficiency and minimize wastage in EC systems.

Eventual Consistency

strong Eventual Consistency (SEC)

Strong Convergence Correct replicas that have delivered the same updates have equivalent state
EventualConsitency \wedge StrongConvergence \equiv SEC

Conflict-free Replicated Data Types (CRDT)

Is an abstract data type designed for replication at multiple nodes with the following properties:
  • Any replica can be modified without coordination with others.
  • Once any two replicas have received the same updates they must share the same state.
CRDTs use commutative operations to ensure strong eventual consistency, eliminating the need for synchronization and achieving high availability and low latency.

Availability + Partition Tolerance + SEC

Operation-based CRDTs

State-based CRDTs

defmodule GSet do
  defstruct elements: MapSet.new()

  def new(), do: %GSet{}

  def add(set, element) do
    %GSet{set | elements: MapSet.put(set.elements, element)}
  end

  def lookup(set, element) do
    MapSet.member?(set.elements, element)
  end

  def compare(set1, set2) do
    MapSet.subset?(set1.elements, set2.elements)
  end

  def merge(set1, set2) do
    %GSet{elements: MapSet.union(set1.elements, set2.elements)}
  end
end

state
based
CRDT

defmodule CounterCRDT do
  defstruct value: 0

  def new(), do: %CounterCRDT{}

  def increment(counter, amount \\ 1) do
    %CounterCRDT{counter | value: counter.value + amount}
  end

  def decrement(counter, amount \\ 1) when counter.value 
		>= amount do
    %CounterCRDT{counter | value: counter.value - amount}
  end
end

Operation
based
CRDT

Complex Crdts

Crdts challenges

  • Interleavings 
  • Byzantine Faults
  • ...

INTERLEAVING ANOMALIES

BYZANTINE FAULT

Syncordian: A Byzantine Fault Tolerant CRDT without Interleaving anomalies

crtds

By Mateo Sanabria Ardila

crtds

CRDTs

  • 17

More from Mateo Sanabria Ardila