Wednesday, February 14, 2018

Let us call them spectre compilers

Though Intel will freak and prefer them called pipeline compilers.

They compile a subset of javascript, or solidify, and used to represent a complete protocol spanning tree, and generate the exact mapping to the pipeline with controlled kernel thread.

The protocols ae spanning tree grammars with known and stable exits, including time out and count out.  The trees can be chained and will inherit global restrictions, and independent protocols can be stacked.  The kernel housekeeping chores are a known code and included in the pipeline format, and that format will be publicly known and exact for any given protocol/processor pair. Variables and secrets in the pipeline are kept anonymous in validation checks. protocols free to call for miner validation at any point, for a fee.

So, rom yh miners perspective, they need only tack the existing stack of protocols in the pipeline, kernel house keeping code deliver a crc tree for use by the iners, and that tree is code in the pipeline at the operation. Protocol graphs may be deleted upon completion, and the crc tree regenerated.

The kernel keeps the secret key hidden, and the miners have the public key for the pipeline crc checks. The beauty of this system is that the customer pays for as much crypto security as it wants.

Validation is a request for queued services, as I pointed out.  The protocol request a validation, and he request goes out to miners.  The next say, eight miners who validate will cause the pipeline to interrupt and existing kernel code in the pipeline 'crcs' both the kernel code and the protocol.  The eight iners either agree, and the external state of the crc tree is set, deleted, or continued.  All other miners accept.

The protocol can offer a low price, wait quite a long time for the miners to validate, or pay a high price.  But protocols have an inheritable timeout and count out which ternates. Pausing the protocol should have no effect on the kernel thread, except via the interrupt channel. 

So, spectre compilers insure, a priori, that the protocols are closed between the ,miners and the kernel thread.

What about look ahead and parallel ALU hardware in the pipeline?  This should not hurt, the pipeline would execute parallelism based on the stacked protocols, but the protocols are absolutely, unconditionally segmented between users.

No comments: