Write Broadcast Protocol
A snoopy cache-coherence action where a write to a shared line is broadcast over the bus so every other cache with that line updates its copy in place, rather than invalidating it. The alternative, and far more common, action is write-invalidate.
Why?
Under write-invalidate, the next time another core reads the line it suffers a cache miss and has to refetch. Write-broadcast avoids that follow-up miss by pushing the new value to every cache up front, so readers stay hot.
Trade-off:
- Pros: no follow-up miss after a write; other cores see the new value immediately in their cache
- Cons: consumes a lot of bus bandwidth, one broadcast per shared write, even for values no other core ever reads again
- For most workloads the bandwidth cost outweighs the miss savings, which is why hardware designers overwhelmingly pick invalidate
Name
On a write, the cache broadcasts the new value out over the shared bus to every other cache. Same origin as radio broadcast: one sender, all snoopers receive.
Not a choice users get to make, the protocol is baked into the CPU’s cache controller.