Workload Utilization Metrics

Draft

Motivation

Tenants/ users of Akash expect to be able to see what amount of allocated resources are being used by their workloads so that they can better manage peak load and also optimize cost/ spend

Summary

This AEP will likely require building the necessary contructs (metrics server/ agent) for collecting utilization metrics from the tenant containers and reporting them through an API that can be quried and graphed for display in clients like Console. The metrics collected initially will likely be GPU (VRAM), CPU, Memeory and Storage.

Estimated completion: 7/15/2026

Created: 12/1/2024

Last Updated: 7/30/2025

Category: Interface

Status: Draft

View next aep

Completion Date: 7/17/2026

This AEP defines how tenants request confidential computing workloads on Akash Network and how providers advertise confidential compute capabilities. Tenants set `params.tee` in their SDL to either `cpu` or `cpu-gpu` to request confidential compute. The provider determines the actual TEE platform (AMD SEV-SNP or Intel TDX) at deployment time based on its hardware. Workloads run inside Kata Containers (micro-VMs) on TEE-capable providers. The spec covers CPU-only and GPU confidential computing, NVIDIA GPU passthrough to Kata VMs, attestation sidecars, and combined CPU+GPU attestation.

Experience the Supercloud.