View synchronisation is an important component of many modern Byzantine Fault Tolerant State Machine Replication (SMR) systems in the partial synchrony model. Roughly, the efficiency of view synchronisation is measured as the word complexity and latency required for moving from being synchronised in a view of one correct leader to being synchronised in the view of the next correct leader. The efficiency of view synchronisation has emerged as a major bottleneck in the efficiency of SMR systems as a whole. A key question remained open: Do there exist view synchronisation protocols with asymptotically optimal quadratic worst-case word complexity that also obtain linear message complexity and responsiveness when moving between consecutive correct leaders? We answer this question affirmatively with a new view synchronisation protocol for partial synchrony assuming minimal clock synchronisation, called \emph{Fever}. If $n$ is the number of processors and $t$ is the largest integer $<n/3$, then Fever has resilience $t$, and in all executions with at most $0\leq f\leq t$ Byzantine parties and network delays of at most $\delta \leq \Delta$ after $GST$ (where $f$ and $\delta$ are unknown), Fever has worst-case word complexity $O(fn+n)$ and worst-case latency $O(\Delta f + \delta)$.

翻译：视图同步是部分同步模型下许多现代拜占庭容错状态机复制（SMR）系统的重要组成部分。粗略而言，视图同步的效率通过从一个正确领导者的同步视图切换到下一个正确领导者的同步视图所需的字复杂度和延迟来衡量。视图同步的效率已成为制约整个SMR系统效率的主要瓶颈。一个关键问题始终悬而未决：是否存在一种视图同步协议，既能实现渐近最优的二次最坏情况字复杂度，又能在连续正确领导者之间切换时获得线性消息复杂度和响应性？我们通过一种名为Fever的新型视图同步协议，在假设最小时钟同步的条件下，对部分同步模型给出了肯定回答。设n为处理器数量，t为满足t < n/3的最大整数，则Fever的容错性为t。在所有包含至多0 ≤ f ≤ t个拜占庭参与方且全局稳定时间（GST）后网络延迟δ ≤ Δ的执行中（其中f和δ未知），Fever的最坏情况字复杂度为O(fn + n)，最坏情况延迟为O(Δf + δ)。