SIMD-0674 would make Solana validators register where they are
Serguey Asael Shinder2 min readQuentin Kniep and Roger Wattenhofer have opened SIMD-0674, "Validator Location Registration", in the Solana improvement proposals repository on 29 September. It is the first half of a pair: SIMD-0675, a geo-aware leader schedule, was opened the same day as a placeholder with its text still to come.
The goal is to order the leader schedule so that consecutive leaders tend to be geographically close, cutting the latency of each handoff. For that every client needs the same location for each validator and the same distance measure. The proposal stores the location as three signed 32-bit integers, WGS84 Earth-centred coordinates in metres, in a new field of the vote account, set with a new vote-program instruction signed by the validator identity. The check that a point lies on the Earth's surface, within about 1 km, and the distance between two validators are both computed in integer arithmetic only. The authors rule out floating-point trigonometry because results can differ across libraries and hardware, which consensus cannot tolerate.
A second feature gate would then make registration mandatory. From that point, the draft says, a validator without a registered location "is treated as if it were not staked": excluded from the leader schedule and from voting, with its stake not counted. A new location takes effect in the schedule two epochs later, like a stake change.
The draft is open about what it does not solve: the check verifies geometry, not truth, and a validator can register any point on the surface. Countering location lies is left to the schedule design in SIMD-0675. Validators may round to city level, and the authors argue metre-level coordinates reveal no more than gossip IP addresses already do.
Why it matters
The proposal is a draft, but it would turn location from something the network infers into something each validator declares, with the loss of all stake weight for refusing. The trade-off between faster handoffs and making operators state where their machines are is what the discussion will be about.