3 ms·
IMO a more practical approach would be to use ordinary encryption inside a CPU, so it wouldn't be possible to extract any data without extremely advanced method
by vbezhenar 6y ago
IMO a more practical approach would be to use ordinary encryption inside a CPU, so it wouldn't be possible to extract any data without extremely advanced methods.
Rough description: CPU have secure memory which contains a private key. Also CPU have certificate with corresponding public key. That certificate is signed by Intel.
Hosting provider publishes an API which allows remote user to communicate with that secure CPU, basically just transmitting encrypted stream to and from CPU.
When you're starting your communication, you establish an encrypted link between your machine and target CPU, located in a data center. You can check authenticity of that CPU by checking certificate signature. Then you're sending any code you would like to execute (encrypted by session key). CPU executes that code. It encrypts RAM in process, so it's not possible to read its contents. And sends back some information (again, encrypted) which you're interested in.
So you have computing resources which are protected from anyone except very few who could extract private key from secure chip. I think that this kind of protection should be enough for many uses. And it would not compromise on performance. The only real world issue would be side channel attacks.
- milkey_mouse 6y agoYou're essentially describing AMD's SEV[1] as mentioned in the article. It piggybacks on the memory encryption implemented in the Zen architecture by giving a separate key for each VM. Ostensibly, the host can't interfere or snoop in the VMs, assuming you trust AMD. I'm surprised it hasn't been more widely adopted. > protected from anyone except very few who could extract private key from secure chip The way I understand chip manufacturing it would be hard to diffuse a separate key into each chip. This means it'd only take one of those very few people to extract and leak the key (or cut out the middleman and leak it from inside AMD) to break it for everyone. > The only real world issue would be side channel attacks. This is probably a much bigger problem in practice. Spectre & related attacks have been effective against Intel SGX[2], another "trusted environment" inside a larger system. [1]: https://developer.amd.com/sev/ https://developer.amd.com/sev/ [2]: https://arstechnica.com/gadgets/2018/08/intels-sgx-blown-wide-open-by-you-guessed-it-a-speculative-execution-attack/ https://arstechnica.com/gadgets/2018/08/intels-sgx-blown-wid...
- mtgx 6y agoWhich Google has started using for its confidential computing VMs.
- rhn_mk1 6y ago> That certificate is signed by Intel. > So you have computing resources which are protected from anyone except very few who could extract private key from secure chip. …and Intel. This has obvious problems in who the root of trust should be, as well as putting all the eggs in the same basket. Also, what you propose is I think equivalent to the SEV technology mentioned in the article.
- giomasce 6y agoFor this scheme you need to trust the CPU (meaning its maker, its implementation, etc). The point of FHE is not trusting anybody.