using triples as the data model for blockchain systems

IPVS/AS

Dennis Przytarski

Using Triples as the

Data Model for

Blockchain Systems

Agenda

• Motivation

• Our approach

• Evaluation

• Example

2Dennis Przytarski • Using Triples as the Data Model for Blockchain Systems

Motivation

• Initially designed for cryptocurrencies

• Simple and fixed data model

• Other scenarios

• Automotive

• Full history of a vehicle

• Real Estate

• Record of land titles

• Voting

• Reduce voter fraud

3

Blockchain Systems

Dennis Przytarski • Using Triples as the Data Model for Blockchain Systems

Motivation

Blockchain

Block n-1 Block nBlock n-2Header

TRXN

...

Header

TRXN

...

Header

TRXN

...

Key-value store(Document store)

World State

queries QueryEngine

Transaction

Key Value

... ...

latest application state

Led

ger

key value

CAR0 {color:“blue“,make:“Ford“}

Exam

ple

is derived from

• Key-value data model

• Simple query engine• World state

• Scenarios

• Transportation/Trucking

• Tracking journey stops,parcel service

• Supply Chain Integrity

• Food chain,waste management

• Requirements

• Flexible information model

• Query engine for

• World state

• History (analytics, audit trails)

4

Blockchain Systems


based on [1]

BLOCKCHAIN SYSTEMS

• immutability

• tamper-resistance

5

DATABASE SYSTEMS

• generic but flexible data model

• powerful query engine

Merge Blockchain and Database Systems

• maintain a flexible information model and

an efficient data representation

• support a powerful query engine

• preserve the integrity of the blockchain‘s

data structure (tamper-resistance)

DESIGN REQUIREMENTS


Data Model

• Key-value

• values are often serialized

• if key is not known, a full scan is necessary

• no powerful query engine, just get/set operations

• Relational

• too strict for immutable data

• schemas evolve over time leading to schema changes

• Triples

• flexible schemas without maintaining an one-size-fits-all schema

• triples are facts data of interest are easier to extract


Merkle B-Tree

7

[2]: The Merkle B-Tree: Li, Feifei, et al. "Dynamic authenticated index structures for outsourced databases." SIGMOD 2006.

Pointer Pointer PointerHash B Hash C Hash D

Data Data Data

Hash A

…Root node

Node

• Pointer: Memory address

• Hash: Hash over the node’s data


Hash B Hash C Hash D’

Architecture

Blockchain

Block n-1 Block nBlock n-2Header

TRXN

...

Header

TRXN

...

Header

TRXN

...


World State

Block n-1 Block nBlock n-2

queries QueryEngine

QueryEngine

Transaction

Key Value

... ...


querieslatest

applicationstate

Merkle B-tree

<Triple>

...

Led

ger

key value

CAR0 {color:“blue“,make:“Ford“}

entity attribute value

CAR0 car/color blue

CAR0 car/make Ford

Exam

ple

Exam

ple

querieshistorical

applicationstateis derived from

Header HeaderHeader


Blockchain


Advantage (1/2)

Replace three data stores by one

Blockchain

Block n-1 Block nBlock n-2Block n-3


World State

Graph databaseRelational database

Analytics

isexported

from

isderived

from

Transaction logon file system

Blockchain

Ledger

Header

TRXN

...

Header

TRXN

...

Header

TRXN

...

Header

TRXN

...

replaced by

Storage Architecture

Triplestore

Blockchain


• The database systems ‚Analytics‘ and ‚World State‘ are not aware of the blockchain‘sdata structure data integrity checks must be done manually

Advantage (2/2)

Built-in Data Integrity Checks (Tamper-resistance)

10

if neededdata integritychecks aredone manually

if neededdata integrity

checks aredone manually

built-in dataintegrity checks

(tamper-resistance)

QueryEngine

QueryEngine

QueryEngine


World State

Graph databaseRelational database

Analytics

isexported

from

isderived

from

Transaction logon file system

Blockchain

Triplestore

Blockchain

Ledger


Problem

• High storage requirements

• Optimization mechanisms ( efficient storage techniques)

• data compression

• data encoding

• reuse of already stored data structures


Example

12

<B9F, name, „Pluto“>

…<B9F, type, „planet“>

…

AEVA <= name

Transaction 1<B9F, name, „Pluto“><B9F, type, „planet“>

Hash FC8

Hash E87

Block 1

<B9F, name, „Pluto“> <B9F, type, „dwarf_planet“>

… …

AEVA <= name

Transaction 2<B9F, type, „dwarf_planet“>

Hash 2F7

Hash E87 Hash C29

Block 2

1

2

4

2 5

QuerySELECT ?type ASOF nWHERE[?object name „Pluto“][?object type ?type]

Node

Key Value

FC8

E87

FAA

2F7

C29

1

2

3

4

5

Nodes arestored in a

key-value store

afterBlock 1

afterBlock 2

Query resultfor n=1: ?type is planetfor n=2: ?type is dwarf_planet

Root ofsecond block

identical node

Hash FAA3

The query engine uses the Merkle B-trees of a block to compute the result of a query

Root offirst block


Contact Details

Thank you!

University of Stuttgart

IPVS/AS

Dennis Przytarski

Email: [email protected]


References

• [1]: Androulaki, Elli, et al. "Hyperledger fabric: a distributed operating system for

permissioned blockchains." Proceedings of the Thirteenth EuroSys Conference. ACM,

2018.

• [2]: Li, Feifei, et al. "Dynamic authenticated index structures for outsourced databases."

Proceedings of the 2006 ACM SIGMOD international conference on Management of data.

ACM, 2006.

• Merkle, Ralph C. "A digital signature based on a conventional encryption function."

Conference on the theory and application of cryptographic techniques. Springer, Berlin,

Heidelberg, 1987.

• C. Mohan, Tutorial: Blockchains and Databases (VLDB 2017)


using triples as the data model for blockchain systems

Documents