Design Details

[cuteolaf]

I'm going to write down how the overall Datagen project works.

You can read the whole application document here.

Key points from the application document

• B-Datagray’s Datagen project concerns the development of a decentralized infrastructure for CPU/GPU cloud computing, in chain, through different blockchain components. The Datagen infrastructure requires the creation of a Substrate-based blockchain with some key features: low latency time, high efficiency (compatibly with the decentralized nature of the network), high degree of decentralization (higher than a traditional PoS would allow) of the physical hardware providing the cloud computing (therefore requiring a customized consensus protocol) and in-chain (or near-in-chain) computation (no off-chain based solutions).
• The actual cloud computing will happen in our FB Datagen (substrate based).
• The validator in the FB must solve the raw computational request and turn it into a hash, within the blocktime, so, the hashed data must propagate in the whole FB, but there is no need for the raw data to be processed multiple times (like in a PoW) in the FB.
• While, what happens in the FB, will be asynchronously and randomly verified by the HB Parachain.

Technical Design

How mathematical works are performed in FB

A user will be paired with an FB validator in some criteria(e.g. geographical location) and request computational works.

The paired FB validator node should

• read & parse raw data
• perform the computation
• save the result
• submit the result to the heavy blockchain

All of the above will be done in an off-chain worker.

Representing & Storing a mathematical problem

• A mathematical problem will be represented as a JSON file stored on decentralized storage systems like IPFS.
• A user will pass the IPFS hash of a problem to FB validators.

Off-chain worker design

Off-chain worker is the core component of an FB node.

It will

• Read problem data from IPFS
• Parse JSON
• Do the computation
• Upload the result to IPFS and get hash
• Calculate hash of the result
• Submit a computation record to HB
  The record will contain
  • problem_hash (IPFS hash of the problem file)
  • result_hash (IPFS hash of the result file)
  • result (hash of the result e.g. H256)

Extrinsics of FB

[Lord-Nymphis]

As for the Milestone 2 of the Web3 foundation grant, there are some simplifications from the functional Datagen infra:

-The infrastructure (both Heavy and Fast Blockchain) is in local;

-The computational problem is simulated by the pallet_computational_work (that we developed in the first milestone of our Grant), which is asking to FB validators to compute the Fibonacci sequence;

-The mathematical problem is stored in the HB runtime (vs IPFS node in the functional infra);

-The user's request to FB validators to run the pallet_computational_work could be simply triggered by a function run in the FB runtime (given that the infra is still in local);

-Parity's Repo https://github.com/paritytech/solo-para-bridge-poc can be used as foundation for the bridge architecture (given that RPC are at their core not so different from off-chain workers / bridges)