The internet today has seen a blast of cloud-based services that has an increasing demand for fast access and short response time, a.k.a. low latency requirements. These services include online gaming, cloud computing, online trading and monetary transactions, video conferencing, VR/AR, etc. To cable networks, they appear as Over-the-Top (OTT) applications that are not managed by operators. Therefore, they cannot be serviced with higher priority like the MSO-offered voice or video. This implies that the cable systems need to provide low latency not only to MSO-managed QoS services, but also to all best effort services just the same. In the past, DOCSIS cable is known for large latency as compared with competition technologies such as FTTP, especially the media access latency in the upstream direction. In D3.1, Low Latency DOCSIS (LLD) was proposed to alleviate the problem. This includes Proactive Grant Service (PGS) that potentially reduces request-grant cycle by 2/3. Despite the advantages, to this day the authors have not seen large deployment of this mechanism, possibly and understandably due to the fact that the current sub-split and mid-split plants still have limited upstream capacity, and thus favor bandwidth utilization over low latency. The introduction of high-split and ultra-split DOCSIS pushes the equilibrium to a new level: with the advent of additional upstream bandwidth that is several times larger than the current, it is affordable to pre-allocate some grants prior to any requests so as achieve a smaller access latency. Similarly, it is also affordable to allocate some more grants on top of the requested bytes. In other words, it does not require CMTS to use accurate grant counts to match the requests. On the other hand, ultra-split DOCSIS as defined in D4.0 requires Minimum Grant Bandwidth (MGB) to be enforced for any transmitting CM at any time of its transmission across the entire extended spectrum, whenever the extended spectrum exceeds 192 MHz. This requirement imposes new challenges to upstream scheduling, including that of PGS. The paper assumes D4.0 Frequency Division Duplex (FDD) is adopted, yet most of the discussions and solutions apply to Full Duplex (FDX) DOCSIS as well. In this paper we show that both latency and data speed of user applications’ might be significantly impacted due to the requirements of MGB. We further discuss various methods in D4.0 upstream scheduling that tackle the MGB issue, and how PGS could be realized using these methods. Thus, it is expected to achieve an optimal latency as well as the desired data speed in operators’ D4.0 networks.