Stratum Mining Channel

A Stratum mining channel is the logical pipe over which a Stratum V2 connection carries jobs and shares between a downstream miner and an upstream server. A single encrypted transport connection can multiplex many channels, each with its own channel ID, target, and extranonce space, and the protocol defines three types — standard, extended, and group — each suited to a different point in the mining topology. Knowing which channel type a device opens tells you how much work it constructs locally versus how much an upstream node does on its behalf, which is the core architectural difference between Stratum V2 and the JSON free-for-all of Stratum V1.

Opening a channel

After the encrypted session is established and SetupConnection succeeds, an end device sends OpenStandardMiningChannel carrying a request ID, its worker identity, its nominal hashrate, and the maximum share target it will accept. The server's success reply assigns the channel_id, an initial share target, a pool-chosen extranonce_prefix, and the group channel the new channel belongs to. Proxies and work-generating nodes instead send OpenExtendedMiningChannel, which adds a min_extranonce_size field so the upstream knows how many extranonce bytes the proxy needs to subdivide among its own downstream devices.

Standard, extended, and group channels

A standard channel is the lightweight path for ASICs. The pool sends NewMiningJob with a complete, ready-to-mine merkle root; the device may only vary the nonce, nTime, and version fields (the last via version rolling). No coinbase construction happens on the machine, which keeps firmware simple — this restriction is what the spec calls header-only mining. An extended channel receives NewExtendedMiningJob instead, carrying coinbase_tx_prefix, coinbase_tx_suffix, and the merkle path, so the node can assemble coinbases and generate work locally without pool round-trips; notably, the coinbase halves arrive with SegWit witness data already stripped, because merkle root calculation needs the stripped txid. A group channel bundles many channels so a single job broadcast and one SetNewPrevHash can address them all at once, slashing bandwidth for proxies that front hundreds of machines.

Shares ride the same pipe

Results travel back on the channel that issued the work. Standard channels use SubmitSharesStandard (channel ID, sequence number, job ID, nonce, nTime, and the full rolled version); extended channels use SubmitSharesExtended, which adds the miner's extranonce contribution so the server can reconstruct the exact coinbase. The server answers with batched success acknowledgements or per-share errors, and it can retune difficulty per channel at any time with SetTarget. Because jobs can be sent as "future" work and activated the instant SetNewPrevHash arrives, a well-implemented channel starts hashing on a new block with essentially zero idle time.

For home miners the practical takeaway: your firmware determines the channel type you can speak. Native Stratum V2 support today means BraiinsOS+ on industrial Antminers and AxeOS/ESP-Miner v2.14.0+ on the Bitaxe; everything else reaches SV2 pools through a translating proxy that opens an extended channel upstream on your behalf. The entire channel scheme runs inside the encryption established by the Noise protocol handshake.

Channel lifecycle and housekeeping

Channels are living objects, not one-shot negotiations. A downstream device that changes its hashrate — a fan failure forcing a downclock, or an autotuner settling on a new operating point — sends UpdateChannel so the server can re-target share difficulty for that channel alone; a mismatch here shows up as a share rate that is uselessly sparse or wastefully chatty. CloseChannel tears down a single channel without touching the transport, so a proxy can retire one dead machine while hundreds of siblings keep hashing on the same connection. The server can reassign extranonce space on the fly with SetExtranoncePrefix, and the sequence numbers on submitted shares exist so a single success message can acknowledge a whole run of shares at once — another of SV2's bandwidth economies. None of this bookkeeping exists in Stratum V1, where one TCP connection means one worker and everything is renegotiated by reconnecting.