Exploiting Caching for Millimeter-Wave TCP Networks: Gain Analysis and Practical Design

Millimeter-wave (mmWave) communications are a key enabler of gigabit access using large spectrum resources. Most relevant works focus on the design of radio access networks (RANs), but a transport layer design tends to be overlooked despite its importance as a data supply to the RAN. Recent studies reported that transmission control protocol (TCP)-the de facto standard for the transport layer-is problematic in mm-Wave communications because the sending rate cannot be controlled due to the drastic channel status change between line-of-sight (LoS) and non-LoS (NLoS); this is known as TCP performance collapse problem (TPCP). The TPCP can be alleviated if a cache is deployed in a base station, such that TCP packets from servers are stored in the cache during NLoS and forwarded to the RAN immediately when the channel becomes LoS. This paper overcomes the TPCP by making full use of the mmWave spectrum via efficient cache utilization. Here, a caching gain is analyzed and defined as the end-to-end rate ratio between TCPs with and without cache. This results show that the gain is maximized by forwarding packets to the RAN whenever the wireless channel is available. Based on this, we propose a novel cache-enabled TCP framework called mmWave TCP (mmTCP) with two key functionalities. The first is a batch retransmission, where packets likely to be lost during NLoS are simultaneously retransmitted once the channel becomes LoS. The second is an online cache management, where the cache is being reallocated to different users given the current channel and cache status. The performance of the mmTCP is evaluated by a realistic network simulator-version 3 under different environments, such that 48% and 10% end-to-end rate improvements are made in singleand multi-user cases, respectively.