Edge Cloud Augmentation for Virtual Reality MMOGs
With the popularity of Massively Multiplayer Online Games (MMOGs) and Virtual Reality (VR) technologies, VR-MMOGs are developing quickly, demanding ever faster gaming interactions and image rendering. We identify three main challenges of VR-MMOGs: (1) a stringent latency requirement for frequent local view change responses, (2) a high bandwidth requirement for constant refreshing, and (3) a large scale requirement for a large number of simultaneous players. Understanding that a cloud-centric gaming architecture may struggle to deliver the latency/bandwidth requirements, the game development community is attempting to leverage edge cloud computing. The objective of this work is to distribute the workflow of VR MMOGs to mobile devices, edge clouds as well central cloud, and they can jointly work to provide ultra-low latency for global players.
To address the challenges, we take a closer look at the game flows in VR-MMOGs and discover that player-initiated events can generally be classified into two categories based on the tolerance levels of response latency. The response to the user's local view change events (which has effect only on his/her screen, eg, mouse movements, map scrolls, selection of a game object without changing it) has much more stringent timeliness requirements compared to the response to the game events (which involves global game state updates, eg, updated scores, bleeding on shot targets). In VR-MMOGs, view change events occur much more frequently than in non-VR-MMOGs because the orientation of the VR device constantly changes and requires immediate (20ms) feedback on the screen. In game events, on the other hand, players can tolerate more than 100ms latency, and in some games this value can be as large as 1 second. Based on the fundamental differences between view change and game events, we believe that they should be treated differently in order to provide the best VR-MMOG user experience.
Firstly, bypassing the center cloud when dealing with view change events can greatly shorten their response latencies, making it possible to have immediate local view updates. Secondly, by rendering frames on edge, we can harness their low latency and high bandwidth. Thirdly, the core network traffic can be largely reduced due to the adoption of edge clouds and the possibility of multicasting game updates to users.
Traditional client-centric gaming fares well when users are equipped with desktops with powerful GPUs (with an average rendering latency pf 10ms). In this case, the view change response latency is rather low, 20ms. The aggregate network traffic is also low 400Mb since only small game update packets are exchanged between the server and the client, with multicast support. Finally, it can obtain a refreshing rate of 60fps. When users with mobile devices (GPUs can render 5 frames per second) try to adopt traditional gaming, the rendering performance degrades significantly. The average view change rendering latency is 300ms, and the average refreshing rate is just around 5 fps. Client-centric video stream gaming faces performance bottleneck in the core network, since it has to unicast frames to each client (which consumes more than 1Tb core network traffic). The frame drop rate is quite high, and therefore the actual frame rate at the client side is only around 7 fps. To alleviate the frequent frame drops, we adopt forward error correction (FEC) so that each frame contains all the events that arrive before the frame is rendered. However, even with this technique, the rendering latency is still high (>1s for both view changes, and game events) since the events can only be delivered by the next frame that is successfully delivered. We note that the performance can be even worse in the real world due to the multiplexing on the back haul links. The proposed VR MMOG architecture EC+ can provide event update latency (< 30ms for view change events and around 100ms for game events) and refresh rate (of 56fps) similar to traditional gaming (desktop), since the renderers in edge clouds are powerful and are close enough to the clients. With such a low update latency on the non-game events, our architecture can have good support on VR applications where users need immediate feedback for the non-game events (eg, look into another direction). While our solution does consume more traffic (>10Tb) in the edge network compared to the traditional gaming, we argue that it is feasible and stable since the ISPs usually have full control of the edge network to ensure QoS.