10.12.25 (inclusive)

monetary

MVP

ICT tasks

Media sphere

IT

Mobile app

The CAP theorem for distributed storage states that in conditions of possible network partitions, it is impossible to simultaneously guarantee both strict consistency and full availability. If the connection between the nodes is broken, you have to choose: either block operations for the sake of consistency, or continue accepting requests at the cost of time differences. The customer's requirement does not allow for either, that is, it goes beyond the provably achievable. The Fisher–Lynch–Paterson result shows that in a fully asynchronous system with the possibility of failure of at least one node, there is no deterministic consensus algorithm that simultaneously guarantees security and survivability. In other words, with an unfavorable message schedule, the algorithm can "fluctuate" indefinitely without reaching a solution. The customer's requirement essentially prescribes the impossible: to ensure guaranteed and prompt decision-making in all possible network scenarios.

The task of "making sure that in case of any network failures and partitions, global storage remains both strictly consistent and always accessible, while meeting a strict latency limit" theoretically has no general solution. Real systems are forced to choose a compromise: sacrifice part of accessibility, weaken the consistency model, or abandon global consensus in favor of local solutions.

Sergeev I.A.