On Correctness of Dynamic Protocol Update

Abstract

Abstract. Replacing or adding network protocols at runtime is problematic -it must involve synchronization of the protocol switch with ongoing local and network communication. We define a formal mathematical model of dynamic protocol update (DPU) and use it to define two DPU algorithms. The algorithms are based on fully-synchronized and lazy strategies. The two strategies implement updates with respectively, strong and weak safety properties. Our model allowed us to express the properties and the DPU algorithms clearly and abstractly, aiding algorithm design and correctness proofs.

Adaptive replication control based on consensus

“…The literature brings some important research work on dynamic protocol update, notably [28,24,6]. These papers refer to the dynamic replacement of modules in general communication protocol stacks.…”

“…These papers refer to the dynamic replacement of modules in general communication protocol stacks. In [28], the authors formalize in great detail weak and strong properties that dynamic protocol update (or DPU for short) has to preserve. It also sketches a DPU protocol based on totally ordered broadcast, which preserves the order of the messages sent by adaptive protocol stacks (by respecting strong DPU properties) , as well as a lazy algorithm that may not preserve delivery order (that respects weak DPU properties).…”

“…In [28], for example, three atomic broadcasts are sequentially executed in order to update a distributed protocol, respecting strong properties. The atomic broadcast protocol update designed in [24] also requires two additional atomic broadcast executions.…”

“…Consensus decision (lines [16][17][18][19][20][21][22][23][24][25][26][27][28][29]. When a consensus decision is reached in a node, the decided value, V Set, is stored into a decided set.…”

“…They rely on strong synchronization on the execution of the update mechanisms and the protocols that are updated, aiming at achieving no interference from the updates on the protocol executions. They define synchronization barriers [6], or assume several steps of totally ordered broadcasts [28,24], so that the system could be able to detect quiescent states (in terms of messages sent and not yet received) before an update occurs. Accordingly, they succeed to have updates that are "isolated" from the protocol stacks execution, so that, the protocols that are updated are not affected by the update process during their execution.…”