Closed synctext closed 4 years ago
ACM publication related work: decentralized blockchain-based electronic marketplaces. Discussion at YCombinator.
Very naive flow diagram, real inter-continental trade lanes are extremely difficult to track. Which person do you trust to track container contents in Singapore, Taiwan or Shanghai? Who is to blame for breakage?
Related work on market for labor and tasks: "This article will provide a theoretical compass to assist in navigating the reinvention of the company. In the same vein, the seismic movements that have fractured the firm, hence redrawing labour law’s borders will be investigated. The exploration is organized as follows. First and foremost, attention will be devoted to the doctrinal debate on the fragmentation of the firm and the decentralisation of the production cycle. " The Invention of the Future. Does ‘Platformisation’ Redefine the Notion of the Firm?
Bitshares live exchange (https://wallet.bitshares.org/#/explorer/blocks):
Really decentral related work? https://github.com/KomodoPlatform/BarterDEX "The Komodo security services are performed by notary nodes"
Docs: http://barterdex.readthedocs.io/en/latest/faq.html#how-to-get-listed-on-barterdex
cost: also called the taker fee. This is about 0.15%
I've listed BarterDEX already in my related work comment. At that time, there was no code. It seems that there's a beta available now. Their technology heavily relies on atomic swaps, a feature that requires BIP65 implemented. While this ensures that exchange of currencies is either executed or not at all, its functionality is limited to trading of cryptocurrencies.
Browsing a bit through their documentation, I found this comparison between state-of-the-art DEX platforms (the last row is quite arbitrary chosen):
I will try to get some more information about their decentralized matching algorithm. Curious to know how they implemented that.
They haven't done any scalability experimentation at all, after reading through their whitepaper. Also, some statements they make, are fuzzy:
With 100 full nodes, thousands of non-relay nodes can be supported, possibly tens of thousands, though that number has not been reached in practice, so we will have to wait and see what the real world limitations are.
... In practice, we are seeing the nearly instantaneous response when all the parameters are properly met.
I compiled BarterDEX and got it running on the DAS5, it's ready for some experiments 👍
Extracted from the IDEX whitepaper:
Due to the competitive nature of the space we are not revealing the exact proposed method for decentralizing the orderbook at this time
Also, according to the whitepaper, their distributed order book will be live in Q3 2018.
Even more fantasy and dreaming from their main project, Aurora (https://auroradao.com/assets/Aurora-Labs-Whitepaper-V0.9.5.pdf):
In addition to guaranteeing open access, operating the infrastructure in a distributed manner brings many additional benefits. [...] Additionally Aurora will be able to access computing power for the best rates possible. It is conceivable that Aurora will eventually incorporate machine learning algorithms operate an AI banker. Operating on a distributed network would give Aurora access to a lot of cheap computing power.
Epic text!
I might have fried the DAS5 cluster at the VU but I managed to reach over 30.000 trades/sec.
The following graph gives the relation between the total number of traders in the network and the average global throughput in trades per second. Each trader trades with another trader at a rate of 20 trades/sec, for a period of 30 seconds. The graph is made after optimizing the code and storing all Trustchain records in memory such that lookups are very efficient.
Also, the graph strongly suggests horizontal scalability.
Repeated the OpenRide experiment, here are two rough plots (still needs some polish and I'm running it more times so I can add some error bars). During the experiment, I differ the broadcast settings, in particular, the fan-out f and TTL. So when using f = 10 and TTL=2, each new ask/bid order reaches 10 + 10^2 = 110 nodes. As we see, lower values of f and TTL lead to suboptimal efficiency, in particular in the situation where f = 5 and TTL = 2. Also, note that increasing the broadcast range further doesn't really improve the matching efficiency anymore.
Average bandwidth usage per network participant:
https://en.m.wikipedia.org/wiki/Myerson–Satterthwaite_theorem "there is no efficient way for two parties to trade a good when they each have secret and probabilistically varying valuations for it, without the risk of forcing one party to trade at a loss."
Don't forget Bitbay #BAY (Bitbay.market which is also a decentralized market with smart contracts.
I now have a trading bot running on the market testnet, which periodically (every 35) creates an order, alternating between an ask and a bid. See here for the latest orders of this node. Note that these blocks are not dual-signed since they are source blocks (and can have any counterparty). I will run this overnight to see whether things work as expected.
Next step: add a second node to the testnet and let these two bots trade.
Nice! This could be the V7.2 key feature. A testnet wallet auto-activated and instant live animation would be cool TorrentFreak promo points. Gigachannel feature will not be fully usable and have gossip updates for 7.2.
After some major refactoring, matchmakers no longer reserve some quantity in their order books when matching. Instead, any conflicts should be resolved between the two trading parties.
The motivation behind this change is that I noticed during my experiments that matchmaking becomes very inefficient if assets are reserved. By avoiding the reservation of assets in the order book, the same order can be matched against multiple other orders and a match
message will be sent to all identities behind the matched orders now. Intuitively, this should lead to better matching efficiency but I should validate this with some experiments on my order datasets.
Lots of related work appearing. As discussed indeed 1 common theme. We need reputations, we are not working on it, and concrete plan is lacking. https://hackernoon.com/dharma-protocol-tokenized-debt-and-funding-through-decentralized-systems-384527f1a75d
Since our market is mostly accountability-oriented, I think a reputation system is a valuable addition and a natural extension of this line of work.
Sorry, to clarify: common theme on these white papers: they have no reputation system, they critically need one, and are delaying work on it. Fundamental science: universal reputation system.
phd thesis outline, outcome of discussion:
After (almost) a week of work, I have the following progress:
venv
with Python 3 and run the AnyDex validation fully with Python 3. Not sure about any improvements regarding speed so far.Next, I will focus on some fundamental issues, namely counterparty fraud detection and improved trade accountability. Hopefully, I can make some progress with the GUI next week, and setup some unit tests for the REST API.
Marketplace with blockchain based reputation system supporting bid,ask, and cancel messages.
Our previous running code, needs work: http://repository.tudelft.nl/islandora/object/uuid:c1816da8-c82e-47bc-bde2-4f35b9a3e775