Proof-of-Archival-Storage

Public Parameters and Values

Protocol Constants

These parameters are fixed at the beginning of the protocol and used by all clients.

SLOT_PROBABILITY: the probability of successful block in a slot (active slots coefficient), currently 1/6. This defines the expected block production rate of 1 block every 6 seconds
INITIAL_SOLUTION_RANGE: solution range for proof-of-replication challenge for the first era
ERA_DURATION_IN_BLOCKS: solution range is adjusted after this many blocks, currently 2016 (as in Bitcoin)
KZG_PUBLIC_PARAMETERS: universal reference values for KZG Polynomial Commitment scheme obtained through an MPC ceremony
KZG_MAX_DEGREE is $2^{15}$ : the maximum degree of the polynomial possible to commit to under the KZG scheme
CONFIRMATION_DEPTH_K: minimal depth after which a block can enter the recorded history (a global constant, as opposed to the client-dependent transaction confirmation depth), currently 100
EXPECTED_VOTES_PER_BLOCK: number of votes expected per block, currently 9
SAFE_BYTES is 31 bytes: the size into which raw history data is partitioned before Archiving for KZG commitment, required to ensure that values fit the size of the point subgroup of BLS12-381 curve
FULL_BYTES is 32 bytes: the size into which encoded data is partitioned after Archiving
NUM_CHUNKS is $2^{15}$ : number of KZG field elements that compose a record, NUM_CHUNKS ≤ KZG_MAX_DEGREE.
NUM_S_BUCKETS: number of s-buckets in a record (and by extension in plot sector), currently $2^{16}$ (NUM_CHUNKS / ERASURE_CODING_RATE)
RAW_RECORD_SIZE is $\lessapprox$ 1 MiB: the size of a record of raw blockchain history before Archiving to be transformed into a piece (SAFE_BYTES * NUM_CHUNKS = 1015808 b)
NUM_RAW_RECORDS is 128: the number of raw records in a Recorded History Segment
RECORDED_HISTORY_SEGMENT_SIZE is roughly 128 MiB: size of raw recorded history segment ready for Archiving (RAW_RECORD_SIZE * NUM_RAW_RECORDS)
RECORD_SIZE is 1 MiB: the size of useful data in one piece (FULL_BYTES * NUM_CHUNKS = 1048576 b)
WITNESS_SIZE is 48 bytes: the size of KZG witness
COMMITMENT_SIZE is 48 bytes: the size of KZG commitment
PIECE_SIZE is $\gtrapprox$ 1 MiB: a piece of blockchain history (RECORD_SIZE + COMMITMENT_SIZE + WITNESS_SIZE = 1048672 b)
ERASURE_CODING_RATE: parameter of erasure code operation, specifies what portion of data is enough to recover it fully in case of loss, currently 1/2
NUM_PIECES is 256: the number of pieces in an Archived History Segment after erasure coding (NUM_RAW_RECORDS/ERASURE_CODING_RATE)
K: space parameter (memory hardness) for proof-of-space, currently 20
RECENT_SEGMENTS: number of latest archived segments that are considered Recent History for the purposes of Plotting (currently 3)
RECENT_HISTORY_FRACTION: fraction of recently archived pieces from the RECENT_SEGMENTS in each sector (currently 1/10)
MIN_SECTOR_LIFETIME: minimum lifetime of a plotted sector, measured in archived segments, currently 4

Dynamic Blockchain Values

These parameters are derived from the blockchain and are dynamically updated over time.

slot_number: current slot number, derived since genesis from the Proof-of-Time chain
solution_range: farming difficulty parameter, dynamically adjusted to control the block generation rate
voting_solution_range: voting difficulty parameter, dynamically adjusted to control the vote generation rate. This range is larger than solution_range to allow for multiple votes per block (solution_range * (EXPECTED_VOTES_PER_BLOCK+1))
global_randomness: used to derive slot challenges, from the Proof-of-Time chain
pieces[]: all pieces of the recorded blockchain history
history_size: number of archived segments of the blockchain history until this moment
segment_commitments[]: hash map of KZG commitments computed over pieces in a segment and stored in the runtime
BlockList: list of public keys that committed offenses

Consensus Parameters

These parameters are fixed at the beginning of the protocol and used by all clients, however they can be updated if necessary.

max_pieces_in_sector: number of pieces to be retrieved from blockchain history by a farmer under Verifiable Sector Construction (VSC) and encoded in a single sector by an honest farmer, currently 1000
sector_size: size of a plotted sector on disk, currently ~1 GiB (max_pieces_in_sector * RECORD_SIZE)

Archiving

Preparing the history

The genesis block is archived as soon as it produced. We extend the encoding of the genesis block with extra pseudorandom data up to RECORDED_HISTORY_SEGMENT_SIZE, such that the very first archived segment can be produced right away, bootstrapping the farming process. This extra data is added to the end of the SCALE-encoded block, so during decoding of the genesis block it'll be ignored.
Once any block produced after genesis becomes CONFIRMATION_DEPTH_K-deep, it is included in the recorded history. Recorded history is added to a buffer of capacity RECORDED_HISTORY_SEGMENT_SIZE.
When added to the buffer, blocks are turned into SegmentItems.
Each segment will have parent segment_header included as the first item. Each segment_header includes hash of the previous segment header and the segment commitment of the previous segment. Together segment headers form a chain that is used for quick and efficient verification that some piece corresponds to the actual archival history of the blockchain.
Segment items are combined into segments. Each segment contains at least two of the following SegmentItems, in this order:
1. The previous segment’s SegmentHeader
2. BlockContinuation, if remained from the previous segment
3. Block(s), as many as fit fully into the current segment, may be none
4. BlockStart, if the next block doesn't fit within the current segment
5. Padding (zero or more) in case padding bytes are necessary to complete the segment to RECORDED_HISTORY_SEGMENT_SIZE
When the buffer (after SCALE-encode) contains enough data to fill a record of RAW_RECORD_SIZE bytes, it is archived:
1. Split the record into $d$ = $2^{15}$ =NUM_CHUNKS chunks of size SAFE_BYTES.
2. Interpolate a polynomial over the chunks of the record with poly(record).
3. Commit to the record polynomial under KZG and obtain the source record_commitment $C_i$ as Commit(polynomial). This will allow us to later compute proofs that a chunk belongs to a particular record.
Once NUM_RAW_RECORDS (128) records have been committed, stack them into a matrix of $n =$ NUM_RAW_RECORDS rows and $d$ columns. Each row is a record.
Erasure code records column-wise with extend(column, ERASURE_CODING_RATE)

This effectively doubles the number of rows and thus, records per segment to NUM_PIECES (256).
Erasure code commitments with extend(record_commitments, ERASURE_CODING_RATE).
Hash each record_commitment, both source and parity, into 254-bit scalar record_commitment_hashes values $h_0…h_{2n}$ .
Interpolate a polynomial segment_polynomial over these hashes and commit to it under KZG to get the segment_commitment $C_s$ as commit(Poly(record_commitment_hashes)).
For each record_commitment_hashes[i] $h_i$ , compute a record_witness $\pi_i$ to the segment_commitment $C_s$ as create_witness(segment_polynomial, i).

This will allow us to prove that a record belongs to an archived history segment without having to provide all other segment commitments at a cost of storing additional 48 bytes per piece.
For each record, form a piece = record || record_commitment || record_witness $(r_i||C_i|| \pi_i)$
Append the segment_commitment to the global segment_commitments[] table of the chain.
The segment now consists of NUM_PIECES records of 1MiB each, NUM_PIECES piece commitments, NUM_PIECES proofs of 48 bytes each and one 48-byte segment_commitment.

The pieces with even indices ((0, 2, …, 254) of this segment) correspond to source records and the pieces with odd indices ((1, 3, …, 255)of this segment) correspond to parity records. The blockchain history data is effectively contained only in pieces with an even piece_index.

Figure 3: Piece Building Process corresponds to steps 5-12 of Archiving.

Plotting

Pre-plotting Initialization

Determining and retrieving the assigned portion of the history

Given the total allocated disk space, reserve some space ( $<2\%$ ) for commitments and other auxiliary information.
Create a single plot of plot_size = allocated_plotting_space. Sizes of farmer plots are independent from size of blockchain history, but must be a multiple of sector_size.
Generate a farmer identity, that is a key pair public_key, secret_key under the digital signature scheme. These signing keys are independent from reward address of an entity (exactly as payout address in Bitcoin mining) described in Block reward address.
Derive an identifier public_key_hash as hash(public_key). This farmer id will also serve as the basis for their single global network peer id in the DHT.
Determine the sector_count = plot_size / sector_size.
Verifiable Sector Construction (VSC):

Determine which pieces are to be downloaded for this sector:
1. Index the sectors sequentially and for each sector derive sector_id = keyed_hash(public_key_hash, sector_index || history_size), where sector_index is the sector index in the plot and history_size is the current history size at the time of sector creation.
2. For each sector, for each piece_offset in 0..max_pieces_in_sector, derive the piece_index in global blockchain history this slot will contain, as follows:
  1. At the start of the chain, if history_size <= RECENT_SEGMENTS / RECENT_HISTORY_FRACTION the pieces for this sector are selected uniformly as piece_index = keyed_hash(piece_offset, sector_id) mod (history_size * NUM_PIECES) for piece_offset in 0..max_pieces_in_sector
  2. Later, when history grows (history_size > RECENT_SEGMENTS / RECENT_HISTORY_FRACTION) to make sure recent archived history is plotted on farmer storage as soon as possible we select RECENT_HISTORY_FRACTION of pieces for each sector from the last RECENT_SEGMENTS archived segments.
  3. For piece_offset in 0..max_pieces_in_sector * RECENT_HISTORY_FRACTION * 2:
    1. If piece_offset is odd, select a piece from recent history as piece_index = keyed_hash(piece_offset, sector_id) mod (RECENT_SEGMENTS * NUM_PIECES) + ((history_size - RECENT_SEGMENTS) * NUM_PIECES)
    2. If piece_offset is even, select a piece uniformly from all history as piece_index = keyed_hash(piece_offset, sector_id) mod (history_size * NUM_PIECES)
  4. The rest of the pieces for this sector are selected uniformly as piece_index = keyed_hash(piece_offset, sector_id) mod (history_size * NUM_PIECES) for piece_offset in (2 * RECENT_HISTORY_FRACTION * max_pieces_in_sector)..max_pieces_in_sector
3. Retain the history_size count at the time of sector creation. This sector will expire at a point in the future as described in Sector Expiration.
Retrieve each piece from the Distributed Storage Network (DSN) and verify against segment commitment obtained from the node.

Figure 4: Verifiable Sector Construction.

For each synced sector, proceed to Plotting phase.

Plotting to Disk

Sealing the history

For each sector, given that all pieces assigned to the sector reside locally in memory, encode pieces row-wise (piece by piece or one piece at a time).

For each piece in the sector:

Extract the bundled record from the piece and retain the record_witness and record_commitment alongside the plot.
Split the extracted record into FULL_BYTES-sized record_chunks, which are guaranteed to contain values that fit into the scalar field.
Erasure code the record with extended_chunks = extend(record_chunks, ERASURE_CODING_RATE)
Derive an evaluation_seed from sector_id and piece_offset within this sector as evaluation_seed = hash(sector_id || piece_offset)
Derive a Chia proof-of-space table pos_table = generate(K, evaluation_seed)
Initialize num_successfully_encoded_chunks = 0 to indicate how many chunks were encoded so far.
Iterate through each (s_bucket, chunk) in extended_chunks:
1. Query the pos_table for a valid proof-of-space proof_of_space for this chunk as find_proof(pos_table, s_bucket).
2. If it exists, encode the current chunk as encode(chunk, hash(proof_of_space)) and increase num_successfully_encoded_chunks by 1. Otherwise, continue to the next chunk.
3. Place the encoded chunk into a corresponding s-bucket for given index s_bucket and set corresponding bit in the encoded_chunks_used bitfield to 1.
If num_successfully_encoded_chunks >= NUM_CHUNKS all chunks were encoded successfully, take the necessary NUM_CHUNKS and proceed to the next record.
Else, if num_successfully_encoded_chunks < NUM_CHUNKS, select extra chunks to store at the end after encoded chunks:
1. Compute num_unproven = NUM_CHUNKS - num_successfully_encoded_chunks
2. Save as extra_chunks the first num_unproven chunks of the source record corresponding to the indices where encoded_chunks_used is not set.
Once all records are encoded, write the sector to disk, one s-bucket at time in order of the index, each followed by encoded_chunks_used bitfield corresponding to selected chunks.
The final plotted sector consists of max_pieces_in_sector many encoded_chunks_used bitfields of size NUM_S_BUCKETS each, NUM_S_BUCKETS s-buckets and commitments and witnesses of pieces in this sector.

Each encoded_chunks_used indicator vector has bit set to 1 at places corresponding to the record positions whose chunks are encoded in this s-bucket and are eligible for farming.

Figure 5: Complete Plotting process for an example sector of 4 pieces

Sector Expiration

After MIN_SECTOR_LIFETIME segments were archived (since sector creation), the farmer should check whether the sector have expired and can no longer be farmed based on the “age” (history size at plotting) of each sector. For each sector sector_id and history_size when this sector was plotted:

When current history size of the chain reaches sector_expiration_check_history_size = history_size + MIN_SECTOR_LIFETIME farmer should check when to expire this sector based on corresponding segment_commitment of last archived segment.
Compute sector_max_lifetime = MIN_SECTOR_LIFETIME + 4 * history_size. With this limitation on sector lifetime, a sector with expire with probability ~50% by the time history doubles since it’s initial plotting point and is guaranteed to expire by the time history doubles again.
Compute expiration_history_size = hash(sector_id || segment_commitment) mod (sector_max_lifetime - sector_expiration_check_history_size) + sector_expiration_check_history_size
When the current history size reaches expiration_history_size, expire this sector.
Replot the sector for new history size as described in Plotting.

Farming

Auditing the history

Challenge Generation

For each slot_number, compute the global_challenge as hash(global_randomness||slot_number). The slot_number and global_randomness are derived from PoT (see derive_global_randomness).
Compute the sector_slot_challenge = global_challenge XOR sector_id
Compute the s_bucket_audit_index = sector_slot_challenge mod NUM_S_BUCKETS

Audit

Read the corresponding s-bucket at s_bucket_audit_index from disk into memory.
For each chunk in the bucket, compute audit_chunk = keyed_hash(sector_slot_challenge, chunk), truncated to 8 bytes
Check if the result (as integer) falls within +/-voting_solution_range/2 of the global_challenge
Sort all the potentially winning audit_chunks by their “quality” - the closest to global_challenge first.
Prove the best chunk and submit the solution.

Proving

For a winning audit_chunk (8 bytes), derive the chunk_offset from the parent full chunk’s position in the s-bucket.
Get the full 32 byte chunk at that chunk_offset.
From encoded_chunks_used bitfields determine if it is an encoded chunk.
If not, chunk is not eligible for proving.
If yes, then determine parent record piece_offset of this chunk.
Compute the steps 1-7 in Recovering the Record and obtain plotted_chunks
Save the proof_of_space for this chunk used for decoding
Recover the extended_chunks_poly using decoded chunks with recover_poly(plotted_chunks).
Retrieve the record_commitment.
Compute the chunk_witness for the decoded winning chunk that ties back to the record_commitment as create_witness(extended_chunks_poly, s_bucket_audit_index), and attach both to the solution for verification against the original record commitment stored in the history.
Retrieve the record_witness that ties back to the segment_commitment.

Produce a solution consisting of

struct Solution {
	public_key:          32 bytes
	reward_address:      32 bytes
	sector_index:         2 bytes
	history_size:         8 bytes
	piece_offset:         2 bytes
	record_commitment:   48 bytes
	record_witness:      48 bytes
	chunk:               32 bytes
	chunk_witness:       48 bytes
	proof_of_space:     160 bytes
}

Submitting a solution

If the winning audit_chunk is within solution_range:

Attach solution to block header.
Forge a new block (as defined in Substrate framework).
Seal the block content with sign(secret_key, block).

Otherwise, submit a vote extrinsic with solution.

Verification

Ensuring a solution comes from a valid plot

If the public_key of the block's farmer is in the block list, ignore the block.
Verify that the consensus log in the block header is correct. This includes solution_range and global_randomness.
Verify that PoT items in the header are correct according to New Blocks
Verify that solution_range and global_randomness are correct for the slot_number of the block.
Compute the global_challenge = hash(global_randomness||slot_number).
Verify that current chain history size in segments is greater than winning piece_index / NUM_PIECES.
Verify that piece_offset ≤ max_pieces_in_sector
Re-derive sector_id
1. Compute public_key_hash = hash(public_key)
2. Re-derive the sector_id = keyed_hash(public_key_hash, sector_index || history_size)
Verify that the sector_id submitted has not expired:
1. Compute sector_expiration_check_history_size = history_size + MIN_SECTOR_LIFETIME and sector_max_lifetime = MIN_SECTOR_LIFETIME + 4 * history_size.
2. Take the archived segment segment_commitment at sector_expiration_check_history_size.
3. Compute expiration_history_size = hash(sector_id||segment_commitment) mod (sector_max_lifetime - sector_expiration_check_history_size) + sector_expiration_check_history_size
4. Check that expiration_history_size is smaller than current history size of the chain.
Re-derive the sector_slot_challenge = global_challenge XOR sector_id
Re-derive the s_bucket_audit_index = sector_slot_challenge mod NUM_S_BUCKETS
Re-derive the evaluation_seed for the record from the sector_id and piece_offset as hash(sector_id || piece_offset)
Verify proof_of_space with Is_Proof_Valid(K, evaluation_seed, s_bucket_audit_index, proof_of_space)
Ensure the chunk satisfies the challenge criteria:
1. Compute the masked_chunk as Encode(chunk, hash(proof_of_space))
2. Compute the keyed hash keyed_hash(sector_slot_challenge, masked_chunk) truncated to audit_chunk size
3. Ensure the result falls within +/- solution_range/2 of the global_challenge
Ensure the provided chunk was correctly extended from a history piece for the assigned record commitment:
1. Verify the chunk_witness for the given chunk, record_commitment, and s_bucket_audit_index
Ensure the encoded record belongs to the assigned slot and that record also belongs to the history:
1. Re-derive the piece_index from the piece_offset and history_size the same way as during pre-plotting (as described in Verifiable Sector Construction)
2. Retrieve the segment_commitment of the segment that contains the assigned piece_index
3. Hash the record_commitment to obtain the record_commitment_hash
4. Verify the record_witness for the piece_index , record_commitment_hash and segment_commitment
Ensure the farmer did not outsource solving to a third-party without revealing their private keys by verifying the farmer_signature with the public_key and chunk.
Verify the signature on the block content.
If another block signed with the solution with same solution (public_key, sector_index, piece_offset,chunk, chunk_audit_index and slot_number) had already been received, report an equivocation by public_key and ignore the block.

The above steps assume standard block and transaction verification.

Extraction

Serving data from the plot

When requested a piece at piece_index identify sector_index and piece_offset where that piece is stored.
For the record plotted at piece_offset in that sector perform record recovery and compose the piece.

Recovering the Record

From the record’s encoded_chunks_used bitfield determine which s-buckets contain chunks of this record.
Read all the plotted_chunks of the parent record from their respective s-buckets. Set the encoded chunk value None in places where encoded_chunks_used is 0. Total length of plotted_chunks should be NUM_CHUNKS/ERASURE_CODING_RATE.
Derive the evaluation_seed for the record from sector_id and piece_offset within this sector as evaluation_seed = hash(sector_id || piece_offset)
Derive a Chia proof-of-space table pos_table = generate(K, evaluation_seed)
Iterate through each (s_bucket, plotted_chunk) in plotted_chunks:
1. If plotted_chunk is None, continue to next chunk.
2. Else, query the pos_table for a valid proof-of-spaceproof_of_space for this chunk as find_proof(pos_table, s_bucket). The proof should exist if the chunk was encoded and plotted.
3. Decode the chunk in-place with decode(plotted_chunk, hash(proof_of_space)).
If the number of plotted_chunks is less then NUM_CHUNKS, we need extra chunks to recover the record. Insert each extra_chunk in places of None in plotted_chunks from the beginning, however many extra chunks there are.
Recover the source record from the decoded chunks with recover(plotted_chunks)).

Composing the Piece

Retrieve the record_commitment and record_witness from the sector table.
Return the recovered piece as record || record_commitment || record_witness.

Randomness Updates

Global randomness is updated every slot with the output of Proof-of-Time chain for that slot. See Proof-of-Time Specification.

Solution Range Updates

Initial solution range INITIAL_SOLUTION_RANGE is computed based on a single plotted sector with the goal of getting a single winning chunk in each slot with probability SLOT_PROBABILITY, as follows:

Let num_audit_chunks be the number of tickets read during single slot audit in a sector. For each piece (max_pieces_in_sector), we audit a single full chunk in a single s-bucket. The probability of s-bucket containing a full chunk at that position is NUM_CHUNKS/NUM_S_BUCKETS (1/2). Thus, a farmer has this many tickets at each slot:

\text{num\_chunks} = \text{max\_pieces\_in\_sector}*\frac{\text{NUM\_CHUNKS}}{\text{NUM\_S\_BUCKETS}}

To compute initial solution range for one sector, maximum possible solution range (U64::MAX) is divided by num_chunks and multiplied by slot probability:

\frac{\text{U64::MAX}}{\text{num\_chunks}}*\text{SLOT\_PROBABILITY}

INITIAL_SOLUTION_RANGE becomes solution_range for the first era.

Global solution_range is adjusted every era accordingly to actual and expected blocks produced per era to keep block production at the same pace while space pledged on the network change:

After every ERA_DURATION_IN_BLOCKS number of blocks, a new era begins.
At the start of a new era, compute and store new solution_range

$\text{next\_solution\_range} = \max\left(\min\left(\frac{\text{era\_slot\_count}}{\text{ERA\_DURATION\_IN\_BLOCKS}}*\text{SLOT\_PROBABILITY}, 4\right), 1/4\right)*\text{solution\_range}$ .

Conversion Rate Between Solution Range and Space Pledged

The relationship between the solution range and the amount of space pledged is dynamically calculated using the function solution_range_to_sectors. This function computes the number of sectors corresponding to a given solution range by adjusting for slot probability and the configuration of data within sectors. Specifically, the maximum possible solution range (SolutionRange::MAX) is first reduced according to the slot probability, which reflects the desired frequency of successful block production. This is further adjusted by the distribution of data, particularly the ratio of the number of chunks per s-bucket in each sector (MAX_PIECES_IN_SECTOR * Record::NUM_CHUNKS / Record::NUM_S_BUCKETS), to account for the probability of an occupied s-bucket being successfully audited in the verification process. The resulting figure is then divided by the current solution range to determine the total number of sectors that this solution range can effectively cover.

const fn solution_range_to_sectors(solution_range: SolutionRange) -> u64 {
    let sectors = SolutionRange::MAX
        // Account for slot probability
        / SLOT_PROBABILITY.1 * SLOT_PROBABILITY.0
        // Now take sector size and probability of hitting occupied s-bucket in sector into account
        / (MAX_PIECES_IN_SECTOR as u64 * Record::NUM_CHUNKS as u64 / Record::NUM_S_BUCKETS as u64);

    // Take solution range into account
    sectors / solution_range
}

Public Parameters and Values​

Protocol Constants​

Dynamic Blockchain Values​

Consensus Parameters​

Archiving​

Plotting​

Pre-plotting Initialization​

Plotting to Disk​

Sector Expiration​

Farming​

Challenge Generation​

Audit​

Proving​

Submitting a solution​

Verification​

Extraction​

Recovering the Record​

Composing the Piece​

Randomness Updates​

Solution Range Updates​

Conversion Rate Between Solution Range and Space Pledged​