What should be the ideal block-size

[ayush-san]

Hi,

I am using juicefs as persistent storage for jupyterhub docker notebooks. Currently, many users have complained that while training the model they have seen performance degradation. Upon debugging I found that juicefs process metrics on that node spiked. I am assuming this is due to small block-size, or can there be another param which I need to tune?

Here's the screenshot of juicefs metrics

[xiaogaozi]

@ayush-san Hi, you could use juicefs profile to analyze which operation is too slow, please refer to the document for more information.