Chronicle Accelerate Building a Digital Currency Peter Lawrey CEO - PowerPoint PPT Presentation

Chronicle Accelerate Building a Digital Currency Peter Lawrey – CEO 19 Apr 2018

Peter Lawrey Java Developer / Consultant for investment banks and hedge funds. 25 years in IT. CEO for 5 years

Major Banks use Chronicle’s framework to build low latency, high throughput trading systems in Java. We believe we can make a difference by bringing HFT to blockchain

5 years old, self funded Core software is open source Projects have over 1000 stars

Agenda • Digital currency economics • Existing solutions • Architecting for performance • A Java model for extension

BTC, XBC - All tokens mined - Reward decreases exponentially - Miners also receive fees

ETH - Initial tokens - New tokens reduce dramatically over time - Fees for mines

IOTA, MIOTA - All preallocated - 2.7 ^ 15 IOTA - Usually expressed as MIOTA (1 million IOTA)

XRP - All preallocated - 60% reserved by Ripple - Might be released over time?

XCL + Fiat & Digital. - Multiple stable currencies - 20% preallocated for early adopters - 40% sold in ICO - 40% reserved for market volatility reduction

How much throughput do we need? VisaNet handles an average of 150 million* transactions every day and is capable of handling more than 24,000 transactions per second. * 1,740/s

How much throughput do we get today? 3 – 7 txn/sec 500 – 1500 txn/sec

How does our block chain perform? Single Server Nodes i7-7820X, 8 core, 32 GB Burst 480,000 trans/sec Sustained 52,000 trans/sec

How does our block chain perform? Multi-server nodes i7-7820X, 8 core, 32 GB Burst millions trans/sec Sustained 400,000 trans/sec

C-like Java framework, Assembly sign/verify, Optimal use of hardware, Core for speed and gateways for integration.

Why use Java? Java brings millions of devs, reasonable speed. C or assembly is harder but faster Code type CPUs Assembly (Ed25519) 40 Operating System (Mostly TCP) 24 Java 4 The interesting code is in Java, but most of the CPU work is in assembly and C

Achieving Consensus Each node runs concurrently as much as possible Periodically they; - Gossip about what has been replicated - Vote on what to include in a round - When the majority vote the same, progress

Blockchain vs distributed ledger When running trusted nodes, you don’t need the overhead of signatures -> distributed ledger (or database) When running untrusted nodes, you need to ensure they don’t act fraudulently -> blockchain.

Fraud Protection Each message is signed is a way that only the private key holder can create. This can be verified using the public key. In our case we use Ed25519 which has a highly optimized 256-bit public/private key and a 64-byte signature.

Optimising for Throughput A key optimization for throughput is batching in to blocks. We use rounds to achieve consensus and blocks increase the number of transactions in each round.

Optimising for Throughput Concurrent Serial Sign/verify Consensus Client TCP Transaction processing

Weekly Checkpointing Replaying from the genesis block doesn’t scale. Ok for 10 trns/sec Not Ok for 100,000 trns/sec We chose to checkpoint weekly NOTE: GDPR requires the ability to be forgotten

Multichain Localize transaction using ISO-3166 Over 4000 regions / sub-chains.

Smarter Sharding If you use hash based sharing, you get fairness however you get no benefit from locality of use. Most transactions occur between parties in a small number of regions (possibly one most of the time)

Smarter Sharding We use ISO 3166 which breaks the world 4000+ regions. The region is at the top of the address when is in base32 e.g. @usny6db2yfdjs – New York, US @auvickjhj4hs3f – Victoria, AUS

Smarter Sharding We start with a single chain for the whole world. As the number of nodes grows, we can split the chain by the highest bit in 2. Each chain can split in 2 etc until we have 4000+ chains

Multichain Grow to 10 to 100 nodes. More nodes means more sub-chains. Aggregate volume to over 100 million per second.

FIX protocol for traditional market integration

AI managed exchanges to reduce volatility

Roadmap Q1 2018 – Working prototype Q2 2018 – increase throughput to 500K/s/sub-chain Test System running. Pre-ICO investment Q3 2018 – ICO Q4 2018 - production

Extending the platform In active development now. Extension is via sub-chains, main chain rarely changes Programming is in Java to make it more accessible to more developers.

Extending the platform Key components to supply; - The Data Transfer Object to hold the transactions. Supports YAML or Binary. - Pre blockchain stage, run on a per client basis. Can handle queries which don’t go to the blockchain. - Post blockchain stage, run by all nodes in the cluster, recorded in the distributed legder.

package cash.xcl.helloworld; public interface HelloWorld { void hello(TextMessage textMessage); }

public class TextMessage extends SignedTextMessage { private long toAddress ; private String text ; public long toAddress() { return toAddress ; } public String text() { return text ; } public TextMessage toAddress( long toAddress) { this . toAddress = toAddress; return this ; } public TextMessage text(String text) { this . text = text; return this ; }

public class HelloWorldGateway implements HelloWorld { private final ErrorMessageLogger client ; private final HelloWorld blockchain ; public HelloWorldGateway(ErrorMessageLogger client, HelloWorld blockchain) { this . client = client; this . blockchain = blockchain; } @Override public void hello(TextMessage textMessage) { if (textMessage.text() == null || textMessage.text().isEmpty()) client .commandFailedEvent( new CommandFailedEvent(textMessage, "text must be set" )); else blockchain .hello(textMessage); } }

public class HelloWorldBlockchainProcessor implements HelloWorld { final MessageLookup<HelloWorld> lookup ; public HelloWorldBlockchainProcessor( MessageLookup<HelloWorld> lookup) { this . lookup = lookup; } @Override public void hello(TextMessage textMessage) { lookup .to(textMessage.toAddress()) .hello(textMessage); } }

@Test public void testMarshalling() { TextMessage tm = new TextMessage( XCLBase32. decode ( "my.friend" ), "Hello World" ); assertEquals ( "!TextMessage {\n" + " sourceAddress: .,\n" + " eventTime: 0,\n" + " toAddress: my.friend,\n" + " text: Hello World\n" + "}\n" , tm.toString()); TextMessage tm2 = Marshallable. fromString (tm.toString()); assertEquals (tm, tm2); }

@Test public void testHelloWorldGateway() { ErrorMessageLogger logger = createMock (ErrorMessageLogger. class ); HelloWorld hello = EasyMock. createMock (HelloWorld. class ); MessageLookup<HelloWorld> lookup = to -> { assertEquals (2002, to); return hello; }; HelloWorldBlockchainProcessor proc = new HelloWorldBlockchainProcessor(lookup); HelloWorldGateway gateway = new HelloWorldGateway(logger, proc); hello.hello( new TextMessage(2002, "Hello World" )); EasyMock. replay (); gateway.hello( new TextMessage(2002, "Hello World" )); EasyMock. verify (); }

Data driven tests hello: { sourceAddress: my.address, eventTime: 0, toAddress: my.friend, text: !!null "" } --- hello: { sourceAddress: my.address, eventTime: 0, toAddress: my.friend, text: "Hello, How are you?" } ---

commandFailedEvent: { sourceAddress: ., eventTime: 0, origSourceAddress: my.address, origEventTime: 0, origProtocol: !!null "" , origMessageType: hello, reason: text must be set } --- hello: { sourceAddress: my.address, eventTime: 0, toAddress: my.friend, text: "Hello, How are you?" } ---

Q & A Blog: http://vanilla-java.github.io/ http://chronicle-accelerate.com enquiries@chronicle-accelerate.com https://github.com/OpenHFT/Chronicle-Accelerate Australia/APAC Contact jerry.shea@chronicle.software