Automated configuration values for high-performance machine types

This document describes how to automatically set default Cloud Storage FUSE values used for high-performance Compute Engine machine types, which are designed to optimize performance for demanding, high-throughput workloads. Values that are manually set at the time of mount will override these defaults.

Machine types

Configuration values are automated for the following high-performance Compute Engine machine types:

Series type Machine type
A2 machine series
a2-megagpu-16g
a2-ultragpu-8g
A3 machine series
a3-edgegpu-8g
a3-highgpu-8g
a3-megagpu-8g
a3-ultragpu-8g
A4 machine series
4-highgpu-8g-lowmem
TPU v5e
ct5l-hightpu-8t
ct5lp-hightpu-8t
TPU v5p
ct5p-hightpu-4t
ct5p-hightpu-4t-tpu
TPU v6e (Trillium)
ct6e-standard-4t
ct6e-standard-4t-tpu
ct6e-standard-8t
ct6e-standard-8t-tpu

Automated configuration values

When a supported machine type is detected, Cloud Storage FUSE automatically applies the following configuration values:

Cloud Storage FUSE configuration file field Cloud Storage FUSE CLI option Automated configuration value
metadata-cache.negative-ttl-secs --metadata-cache-negative-ttl-secs 0
metadata-cache.ttl-secs1 --metadata-cache-ttl-secs1

-1
metadata-cache.stat-cache-max-size-mb --stat-cache-max-size-mb 1024
metadata-cache.type-cache-max-size-mb --type-cache-max-size-mb 128
implicit-dirs --implicit-dirs true
file-system.rename-dir-limit --rename-dir-limit 200000

1Setting this configuration to -1 significantly boosts performance by always serving files from the cache. Be aware that this configuration bypasses consistency checks, which can lead to serving outdated data. For details on managing data consistency, refer to Overview of caching in Cloud Storage FUSE.

Further performance tuning

When you use a high-performance Cloud de Confiance by S3NS machine type, the configuration values detailed on this page are automatically applied. However, you can further fine-tune your machine for optimal performance using the following methods:

  • Use the Performance tuning best practices guide to improve Cloud Storage FUSE using key Cloud Storage FUSE features and configurations to achieve maximum throughput and optimal performance.

  • If you're running training, serving, or checkpointing and Just in Time (JIT) cache workloads on Google Kubernetes Engine clusters that use Cloud GPUs or Cloud TPU to access large datasets in Cloud Storage, you can streamline your setup by utilizing pre-configured YAML files to mount your Cloud Storage buckets directly into your pods more efficiently. For more information and instructions on how to use pre-configured GKE YAML files, see Use pre-configured GKE YAML files to optimize Cloud Storage FUSE performance.

  • If you're running training, serving, or checkpointing workloads using Cloud Storage FUSE, you can use the profile field or --profile command option to automatically adjust specific Cloud Storage FUSE configurations for optimal performance based on the specific workload type. For more information, see Profile-based configurations for AI/ML workloads.

What's next