| `ExtendedMatrix` | `List[Cell, MAX_CELLS_IN_EXTENDED_MATRIX]` | The full data of one-dimensional erasure coding extended blobs (in row major format). |
Each node downloads and custodies a minimum of `CUSTODY_REQUIREMENT` subnets per slot. The particular subnets that the node is required to custody are selected pseudo-randomly (more on this below).
A node *may* choose to custody and serve more than the minimum honesty requirement. Such a node explicitly advertises a number greater than `CUSTODY_REQUIREMENT` via the peer discovery mechanism -- for example, in their ENR (e.g. `custody_subnet_count: 4` if the node custodies `4` subnets each slot) -- up to a `DATA_COLUMN_SIDECAR_SUBNET_COUNT` (i.e. a super-full node).
The particular columns that a node custodies are selected pseudo-randomly as a function (`get_custody_columns`) of the node-id and custody size -- importantly this function can be run by any party as the inputs are all public.
*Note*: increasing the `custody_size` parameter for a given `node_id` extends the returned list (rather than being an entirely new shuffle) such that if `custody_size` is unknown, the default `CUSTODY_REQUIREMENT` will be correct for a subset of the node's custody.
At each slot, a node needs to be able to readily sample from *any* set of columns. To this end, a node SHOULD find and maintain a set of diverse and reliable peers that can regularly satisfy their sampling demands.
A node runs a background peer discovery process, maintaining at least `TARGET_NUMBER_OF_PEERS` of various custody distributions (both `custody_size` and column assignments). The combination of advertised `custody_size` size and public node-id make this readily and publicly accessible.
*Note*: while high-capacity and super-full nodes are high value with respect to satisfying sampling requirements, a node SHOULD maintain a distribution across node capacities as to not centralize the p2p graph too much (in the extreme becomes hub/spoke) and to distribute sampling load better across all nodes.
*Note*: A DHT-based peer discovery mechanism is expected to be utilized in the above. The beacon-chain network currently utilizes discv5 in a similar method as described for finding peers of particular distributions of attestation subnets. Additional peer discovery methods are valuable to integrate (e.g., latent peer discovery via libp2p gossipsub) to add a defense in breadth against one of the discovery methods being attacked.
In this construction, we extend the blobs using a one-dimensional erasure coding extension. The matrix comprises maximum `MAX_BLOBS_PER_BLOCK` rows and fixed `NUMBER_OF_COLUMNS` columns, with each row containing a `Blob` and its corresponding extension. `compute_extended_matrix` demonstrates the relationship between blobs and custom type `ExtendedMatrix`.
For each column -- use `data_column_sidecar_{subnet_id}` subnets, where `subnet_id` can be computed with the `compute_subnet_for_data_column_sidecar(column_index: ColumnIndex)` helper. The sidecars can be computed with the `get_data_column_sidecars(signed_block: SignedBeaconBlock, blobs: Sequence[Blob])` helper.
To custody a particular column, a node joins the respective gossip subnet. Verifiable samples from their respective column are gossiped on the assigned subnet.
At each slot, a node SHOULD select at least `SAMPLES_PER_SLOT` column IDs for sampling. It is recommended to use uniform random selection without replacement based on local randomness. Sampling is considered successful if the node manages to retrieve all selected columns.
Alternatively, a node MAY use LossyDAS selecting more than `SAMPLES_PER_SLOT` columns while allowing some missing, respecting the same target false positive threshold (the probability of successful sampling of an unavailable block) as dictated by `SAMPLES_PER_SLOT`. The table below shows the number of samples and the number of allowed missing columns for this threshold.
A node SHOULD maintain a diverse set of peers for each column and each slot by verifying responsiveness to sample queries.
A node SHOULD query for samples from their peers via `DataColumnSidecarsByRoot` request. A node utilizes `get_custody_columns` helper to determine which peer(s) it could request from. If more candidate peers are found, a node SHOULD randomize it's peer selection to distribute sample query load in the network. Nodes MAY use peer scoring to tune this selection (for example, by using weighted selection or by using a cut-off threshold).
If a node already has a column because of custody, it is not required to send out queries for that column.
If a node has enough good/honest peers across all columns, and the data is being made available, the above procedure has a high chance of success.
Due to the deterministic custody functions, a node knows exactly what a peer should be able to respond to. In the event that a peer does not respond to samples of their custodied rows/columns, a node may downscore or disconnect from a peer.
If a node fails to sample a peer or fails to get a column on the column subnet, a node can utilize the Req/Resp message to query the missing column from other peers.
*Note*: A node always maintains a matrix view of the rows and columns they are following, able to cross-reference and cross-seed in either direction.
*Note*: There are timing considerations to analyze -- at what point does a node consider samples missing and choose to reconstruct and cross-seed.
*Note*: There may be anti-DoS and quality-of-service considerations around how to send samples and consider samples -- is each individual sample a message or are they sent in aggregate forms.
## DAS providers
A DAS provider is a consistently-available-for-DAS-queries, super-full (or high capacity) node. To the p2p, these look just like other nodes but with high advertised capacity, and they should generally be able to be latently found via normal discovery.
DAS providers can also be found out-of-band and configured into a node to connect to directly and prioritize. Nodes can add some set of these to their local configuration for persistent connection to bolster their DAS quality of service.
Such direct peering utilizes a feature supported out of the box today on all nodes and can complement (and reduce attackability and increase quality-of-service) alternative peer discovery mechanisms.
## A note on fork choice
*Fork choice spec TBD, but it will just be a replacement of `is_data_available()` call in Deneb with column sampling instead of full download. Note the `is_data_available(slot_N)` will likely do a `-1` follow distance so that you just need to check the availability of slot `N-1` for slot `N` (starting with the block proposer of `N`).*
The fork choice rule (essentially a DA filter) is *orthogonal to a given DAS design*, other than the efficiency of a particular design impacting it.
In any DAS design, there are probably a few degrees of freedom around timing, acceptability of short-term re-orgs, etc.
For example, the fork choice rule might require validators to do successful DAS on slot `N` to be able to include block of slot `N` in its fork choice. That's the tightest DA filter. But trailing filters are also probably acceptable, knowing that there might be some failures/short re-orgs but that they don't hurt the aggregate security. For example, the rule could be — DAS must be completed for slot N-1 for a child block in N to be included in the fork choice.
Such trailing techniques and their analysis will be valuable for any DAS construction. The question is — can you relax how quickly you need to do DA and in the worst case not confirm unavailable data via attestations/finality, and what impact does it have on short-term re-orgs and fast confirmation rules.
## FAQs
### Row (blob) custody
In the one-dimension construction, a node samples the peers by requesting the whole `DataColumn`. In reconstruction, a node can reconstruct all the blobs by 50% of the columns. Note that nodes can still download the row via `blob_sidecar_{subnet_id}` subnets.
The potential benefits of having row custody could include:
1. Allow for more "natural" distribution of data to consumers -- e.g., roll-ups -- but honestly, they won't know a priori which row their blob is going to be included in in the block, so they would either need to listen to all rows or download a particular row after seeing the block. The former looks just like listening to column [0, N) and the latter is req/resp instead of gossiping.
2. Help with some sort of distributed reconstruction. Those with full rows can compute extensions and seed missing samples to the network. This would either need to be able to send individual points on the gossip or would need some sort of req/resp faculty, potentially similar to an `IHAVEPOINTBITFIELD` and `IWANTSAMPLE`.
However, for simplicity, we don't assign row custody assignments to nodes in the current design.
To start with a simple, stable backbone, for now, we don't shuffle the subnet assignments via the deterministic custody selection helper `get_custody_columns`. However, staggered rotation likely needs to happen on the order of the pruning period to ensure subnets can be utilized for recovery. For example, introducing an `epoch` argument allows the function to maintain stability over many epochs.
