Dedicated Profile of Multi-staged Microservice Application
With ever-growing workloads in the cloud, it is critical to reinforce the capacity of APDCs to meet the scale-out, emerging applications. Particularly, container and container orchestration have recently catalyzed the shift from monolithic applications to massive heterogeneous and loosely-connected microservices. This enables more scalability potentials at a microsecond-level granularity. Meanwhile, this also requires us to re-think the challenges of accommodating convectional power management to microservice’s features in APDCs.
One of the challenges is to match server-level power budget variation to fine-grained microservice heterogeneity.
Current power management frameworks involve hundreds of milliseconds or even seconds of control latency, making it difficult to make informed power allocation decisions for microsecond-level microservices.
Generally, when facing a myriad of short-lived tasks such as microservices, traditional power management needs to be re-examined at different layers to meet the new latency requirements.