Leveraging knowledge-as-a-service (KaaS) for QoS-aware resource management in multi-user video transcoding

TL;DR

This paper tackles the problem of QoS-aware resource management for multiple concurrent applications on shared nodes by introducing Knowledge-as-a-Service (KaaS) built with Reinforcement Learning. The authors formalize the environment as a Markov Decision Process $(\mathcal{S},\mathcal{A},\mathcal{P},\mathcal{R})$ and present methods to learn multiple QoS-specific policies, including decomposing state and reward spaces and offline precomputation of transition probabilities $P'$. They further propose a 3-tier heuristic, 3TierPol, to apply these policies in real time under varying resource constraints, demonstrated on a real HEVC video transcoding setup with regular and premium users. Experimental results show that the approach can adapt to heterogeneous QoS requests and increase the number of concurrently served users by up to $1.24\times$, validating the practicality and scalability of KaaS for QoS-aware resource management in multi-user video transcoding. The work provides a concrete pathway to scalable, automated policy design and deployment in shared computing environments.

Abstract

The coexistence of parallel applications in shared computing nodes, each one featuring different Quality of Service (QoS) requirements, carries out new challenges to improve resource occupation while keeping acceptable rates in terms of QoS. As more application-specific and system-wide metrics are included as QoS dimensions, or under situations in which resource-usage limits are strict, building and serving the most appropriate set of actions (application control knobs and system resource assignment) to concurrent applications in an automatic and optimal fashion becomes mandatory. In this paper, we propose strategies to build and serve this type of knowledge to concurrent applications by leveraging Reinforcement Learning techniques. Taking multi-user video transcoding as a driving example, our experimental results reveal an excellent adaptation of resource and knob management to heterogeneous QoS requests, and increases in the amount of concurrently served users up to 1.24x compared with alternative approaches considering homogeneous QoS requests.

Figures (5)

• Figure 1: Rewards obtained in the different states for different combinations of sub-reward definitions (left) and coefficients (right)
• Figure 2: Sub-reward functions used for the different policies: three sub-rewards for different levels of quality (above), and the functions for real-time encoding and power (below)
• Figure 3: Reward functions defined for each policy
• Figure 4: System behaviour timelines and metrics obtained when encoding the v1 sequence with all different policies for Regular users (left) and Premium users (right). The yellow line indicates real-time encoding (24 FPS)
• Figure 5: Users per minute attended by each approach with users requests arriving every 10 seconds (top), and quality obtained by each approach when encoding videos from premium users (bars), and percentage of time 3TierPol is in the S2 state (line)