Not in detail, but enough to jog my memory. The Intel order book has been an engineers language for me most of the career.
But it should be possible to compile directly to the instruction cache such that none of the look aheads need be called. Thus, if the instruction cache code is known and closed, the pipeline can be verified if the only interrupt is the interrupt associated with the kernel's private key.
The kernel key interrupt only comes from validated notary functions (or timeout). The notary function is publicly observable and validated by consensus. Otherwise the instruction cache takes a known, provable stable path.
What is going on is a temporary kernel switch, the normal kernel code sits in abeyance and yields control of the pipeline to the notaries. Inside the protocols are guaranteed for finite count and timeout exits. As long as key variables are encrypted, the secure and regular kernels can be swapped at stable protocol points.
The proof is that there is no faster path to the atomic swaps. Nothing except hardware failure will prevent the next few steps in the protocol from executing. Miners know how to do his, so do the kernel folks.
Does it slow the network?
No, it is a fee for service it is optimally congested. For most of us doing retail shopping, we only have notary duty about once a month, no one is that concerned about our every day hand held pipelines. For billion dollar swaps, the processor steps through seconds at a time, every notary from here to Bangkok is watching the deal. Typical online vendors will see no change in function, we can do a smooth context switch between pipeline control and asynch switching.
Spectre was a big deal. Spectre exposed the pipeline layer and suddenly we have a complete monetary network layering. Everyone knows their job, especially miners who build the general purpose notary system.
No comments:
Post a Comment