After defining resilience in your system, you must know for yourself that if you want Byzantine behavior from your nodes or not. Or do you just want your system to be crash-fault tolerant!
If you stick to the higher "Byzantine" resilience, you are most likely looking at a PBFT-variant consensus, assuming your system is permissioned. In most cases, the requirement for a "safe" and "live" consensus is 66% "honest" nodes. On the other hand, if you just assume crash-tolerance, then the requirement is 51% "online" nodes.
A real example is Hyperledger, where some of its frameworks are only crash-fault tolerant (e.g. Hyperledger Fabric) but others are also Byzantine-fault tolerant (e.g. Hyperledger Indy).