How Fluid v0.3 Accelerates Kubernetes PVC Access and Enhances Data Security

Overview

Fluid is an open‑source cloud‑native platform for data‑intensive workloads. Version 0.3 (released September 2020) adds three core capabilities: generic storage‑volume acceleration, fine‑grained dataset access control, and default parameter optimizations.

1. Kubernetes volume acceleration

Version 0.3 extends acceleration to both Persistent Volume Claims (PVC) and HostPath mounts. Any storage that can be presented as a Kubernetes PVC or as a node‑local directory can be cached, benefit from data‑affinity scheduling, and achieve higher I/O performance.

Example Dataset custom resource:

apiVersion: data.fluid.io/v1alpha1
kind: Dataset
metadata:
  name: fluid-imagenet
spec:
  mounts:
  - mountPoint: pvc://nfs-imagenet
    name: nfs-imagenet

Performance evaluation using a TensorFlow ResNet‑50 benchmark:

Training time reduced from 2 h 15 m 59 s to 1 h 43 m 43 s

Throughput increased from 15 024 to 20 506 images/second

Top‑5 accuracy changed from 0.9228 to 0.9204

2. Dataset access permission control

Fluid propagates the underlying storage’s POSIX file ownership and mode to the Fluid nodes, preserving the original access‑control policies after deployment. This enables strict data isolation in multi‑tenant machine‑learning platforms.

The “temporary borrowing” feature allows administrators to grant short‑term access to another user’s dataset by converting ownership at the node level, providing finer‑grained and flexible permission management.

Reference documentation (plain URLs):

PVC acceleration guide: https://github.com/fluid-cloudnative/fluid/blob/master/docs/zh/samples/accelerate_pvc.md

HostPath acceleration guide: https://github.com/fluid-cloudnative/fluid/blob/master/docs/zh/samples/hostpath.md

Non‑root access guide: https://github.com/fluid-cloudnative/fluid/blob/master/docs/zh/samples/nonroot_access.md

3. Default parameter optimization

Prior to v0.3 users needed to manually tune many Alluxio and Fuse parameters. The release ships with a curated set of default configurations that deliver strong performance for typical AI and big‑data workloads, reducing operational effort.

Project resources

Source repository: https://github.com/fluid-cloudnative/fluid

Release page: https://github.com/fluid-cloudnative/fluid/releases