The high-performance SSD examples in the docs mostly involve installing dependencies, builds, and Git. My intended tasks are closer to batch spreadsheet processing, image cropping, and file exports.
How would you tell whether storage, computation, or model calls are the bottleneck? Has anyone compared similar workloads? I'd like to know which metrics to check before choosing a higher specification.
I would start with standard storage and time a small, representative batch before paying for a different tier. “Spreadsheets and images” alone do not identify the bottleneck. Buda's storage guide positions standard storage for sequential document, image and data-file reads; high-performance storage is aimed at many small files, random I/O, dependency installs, builds and Git. Image cropping or spreadsheet calculation may be CPU-bound, while AI steps or web requests may be waiting outside the disk entirely.
Have the Agent measure time spent reading/writing files, local processing, and model or network waits on the same sample, alongside CPU/memory in Computer settings. If many small images are repeatedly opened and file I/O dominates, compare the same files and script on a high-performance Agent. If the model or image computation dominates, faster storage may change little. Compare final task time and output quality, not just raw disk speed. The storage tier is chosen when an Agent is created, so you cannot simply toggle an existing Agent in place; high-performance storage belongs to a paid tier.
Roughly how many files and what sizes are in one batch? Is the slow part opening many small files or processing each file after it is loaded? The Chinese discussion covers the same test.
Sign in to join the discussion. Reading is always open; you only need an account to post.
Sign inThe high-performance SSD examples in the docs mostly involve installing dependencies, builds, and Git. My intended tasks are closer to batch spreadsheet processing, image cropping, and file exports.
How would you tell whether storage, computation, or model calls are the bottleneck? Has anyone compared similar workloads? I'd like to know which metrics to check before choosing a higher specification.
I would start with standard storage and time a small, representative batch before paying for a different tier. “Spreadsheets and images” alone do not identify the bottleneck. Buda's storage guide positions standard storage for sequential document, image and data-file reads; high-performance storage is aimed at many small files, random I/O, dependency installs, builds and Git. Image cropping or spreadsheet calculation may be CPU-bound, while AI steps or web requests may be waiting outside the disk entirely.
Have the Agent measure time spent reading/writing files, local processing, and model or network waits on the same sample, alongside CPU/memory in Computer settings. If many small images are repeatedly opened and file I/O dominates, compare the same files and script on a high-performance Agent. If the model or image computation dominates, faster storage may change little. Compare final task time and output quality, not just raw disk speed. The storage tier is chosen when an Agent is created, so you cannot simply toggle an existing Agent in place; high-performance storage belongs to a paid tier.
Roughly how many files and what sizes are in one batch? Is the slow part opening many small files or processing each file after it is loaded? The Chinese discussion covers the same test.
Sign in to join the discussion. Reading is always open; you only need an account to post.
Sign in