Dragonfly Partner: How did I miss the opportunity to invest in Solana seed turbines?

Original author: @hosseeb

Note: On the fifth anniversary of Solana's birth, @hosseeb, partner of Dragonfly Capital, posted a tweet today, reviewing how he missed the Solana seed investment at $0.04 unit in 2018 and missed more than a thousand times the gains. At the same time, the original investment memorandum was attached to express my nostalgia. In addition, we also excerpted the discussions between Solana Toly and Hosseeb under this tweet.

The following are the original details:

I turned down an opportunity to participate in the @solana seed round for $0.04 in early 2018.

At the current price, it is equivalent to missing a return of 3250 times.

Solana was one of the first projects I evaluated as a junior VC. At that time, I was still very cute and naive and confident, and would write memorandums for every project that gave up on investment.

Rereading this memo now is simply "a embarrassing scene for junior VCs". At the time we were all addicted to finding “Ethereum killers”, studying consensus protocols, and what technology would replace EVM/eWASM.

So, this is the original text of the entirely unedited memo – the worst investment in MISS in my career.

Happy birthday, Solana! 🎂

Memorandum content

1. After reading the white paper, my shorthand is as follows:

• Their significant innovation is the Proof of History (PoH). Essentially this is a verifiable time delay function that uses continuous hashing operations similar to sequential proof of work. In other words, a time maintainer is selected that continually hash a value iteratively and publishes all intermediate hash values. Since this process must be performed serially on a single core and cannot be parallelized, nodes should be able to predict the amount of time elapsed between consecutive hashes (presumably based on their knowledge of hardware performance?).

• The PoH node also mixes any current state (such as the transaction to be submitted) into these hashes. This allows you to create event history that can be reliably timestamped.

• If there is a problem with the PoH node or it cannot be guaranteed to be online, they propose a solution to allow multiple PoH nodes to mix states with each other regularly.

• A set of validator nodes replays and verifies the operations of the PoH node (the verification process can be more efficiently parallelized through the MapReduce architecture). These validators agree on using PoS through a Casper-like protocol. If a PoH node is found to have Byzantine problems or behave improperly, the validator node can elect a new PoH node to replace it.

• It looks like they will develop payment and smart contract features.

• They claim to reach 710,000 TPS and achieve 35,000 TPS on a single-node test network.

2. My thoughts:

• Their numbers are totally nonsense. 710,000 TPS is ridiculous; even Google's search volume per second is less than 100,000. This data is placed on the most prominent position on their website, which makes me very alert.

• Recall the comments that the white paper was well written before. High-level content is good, but the technical details are very lacking and vague. As a description of a consensus protocol, rigor is disappointing.

• The team is mainly composed of Qualcomm's underlying engineers. CEOs and CTOs are mainly engaged in operating systems, embedded systems, GPU optimization and compilers. Their background in distributed systems and cryptography is obviously not strong enough, which is clearly manifested in the paper. Byzantine fault tolerance issues are poorly handled. It reminds me of Raiblocks/Nano’s white paper (they are also under-level engineers).

• And the content like this in the white paper makes me doubt:

[Original text of Solana White Paper, Section 5.12]

"PoH allows network validators to observe events and their time with some degree of determinism. When the PoH generator generates a message flow, all validators need to submit their signature of the status within 500ms. This value can be further reduced based on network conditions. Since each verification is input into the stream, everyone in the network can verify that all validators submitted their votes within the specified timeout without directly observing the voting process."

• This is not a consensus protocol. It is quite problematic to assume that limiting 500ms to consensus on messaging is not meaningfully implemented and Byzantine fault tolerance is not meaningfully implemented. Besides, how do they measure 500ms? How do other nodes in the system agree on the 500ms process, considering that they will estimate the time elapse based on the number of iterative hashes performed? Furthermore, how will they resolve the deviation of clock speed over time due to hardware improvements, hardware failures, or noise? The time issue in distributed systems is very complex and I don't think they realize how difficult it is.

• Besides, who cares about time? Is this a big problem in the blockchain field? Are people not satisfied with the granularity of block time of 15 seconds/1 second (like DFINITY or something)? I don't think that's a big deal, and the complexity and confusion they introduced into the protocol doesn't seem to bring much value.

• They have a section that specifically discusses attacks and motivational misalignment issues. Their response to the attack is completely unconvincing and is equally lacking in rigor or detail.

• They have a whole chapter discussing the copy proof, just like Filecoin does. What to do? Tell me your consensus protocol and how to implement transactions and accounts, and what characteristics your blockchain will have. I don't care about data storage proof.

• There is also a big paragraph that begins with describing smart contracts, but only says that they will use LLVM as the backend to support multiple platforms. But nothing else was mentioned.

• A lot of things about GPU and parallelization. This reveals a strange sense of focus – if they need to implement the BFT consensus protocol and the available smart contract platform, they should not be addicted to the parallel processing of their packet format. I remember they were the same in the demos I've seen - spending most of the time discussing how to use these nodes to handle optimizations, with little time actually describing their consensus protocol.

Conclusion: I will never invest in this project

Interestingly, after 5 years, when Haseeb @hosseeb tweeted to wish Solana that she had successfully gained a place in Encryption and joked about how she missed a big opportunity when she was young, Solana Toly @aeyakovenko replied in this tweet : "All your initial worries are indeed reasonable. In essence, this is a bet - betting on whether we can solve these problems while maintaining the underlying advantages that other teams do not have."

Then Haseeb replied to Toly: "I think this is the lesson. Your persistence in underlying optimization and unique attack angles is not available to other teams. This kind of strengths and weaknesses that can be used to the extreme is the most important thing. I didn't realize this at that time."