ckBTC’s Archive Proposal Turns Memory Into a Reliability Budget

The ckBTC archive canister is the subject of a narrowly scoped NNS proposal with a broader operational lesson: memory allocation is not only a capacity setting, but also a reliability budget.

Proposal 143901 seeks to change the archive canister’s memory_allocation from 0, which uses best-effort allocation, to 2,147,483,648 bytes—2 GiB. The target is canister nbsys-saaaa-aaaar-qaaga-cai.

According to the proposal material, the archive currently uses 924 MiB. The requested setting therefore leaves about 1,124 MiB of headroom, or roughly 2.22 times current usage. Block archiving is currently disabled, so present usage is static. The proposal nevertheless selects 2 GiB to preserve room if archiving resumes, reducing the chance that another governance vote will be needed during future subnet pressure.

The distinction between the two allocation modes matters. With best-effort allocation, memory growth above a subnet’s storage-reservation threshold can require cycles to be reserved in advance. Growth can then be rejected when the canister’s reserved-cycle limit is exhausted. With a reserved allocation, growth within the allocation does not require a new reservation and is not rejected on that specific basis.

That predictability has a price. ICP’s developer documentation says a memory allocation is rented over time based on the allocated amount, regardless of actual usage. The proposal estimates roughly 26.2 trillion cycles per GiB per year on the 34-node fiduciary subnet, or about 52.4 trillion cycles annually for the full 2 GiB. Because the archive already pays for memory it uses, the incremental cost is estimated at about 28.7 trillion cycles per year for the unused headroom.

The proposal also says the fiduciary subnet is currently well below its 750 GiB reservation threshold, so setting the allocation is expected to reserve no cycles at present. That condition is important: the same change under storage pressure could require a substantial reservation or be rejected.

For Chain Fusion infrastructure, the useful takeaway is not that every canister should over-allocate memory. It is that long-lived custody and accounting components may need explicit resource guarantees when failure during growth is more costly than a predictable operating charge. The trade-off is straightforward: pay for reserved capacity now to reduce one class of future availability risk.

The live dashboard page confirms the proposal title, but its retrieved rendering did not expose the full payload or voting details; the numerical settings and cost estimates in this article come from the supplied proposal material and should be rechecked before publication.