Score:0

Ubuntu 20.04, SSD utilization is too high with iostat. But the read speed is low

vn flag

I am training deep learning models with pytorch, Ubuntu 20.04. But the io performance with the ssd is strange. When I start training, the training speed is full in whole day. But it sometime slows down the next day and won't speed up no matter how many times I have restarted my program.

I found the io performance maybe the key. Because the size of single item of my data is 76Mb, and each batch have 8 item. I use iostat and get the result below: enter image description here

The %util is always above 90%, whether the training speed is full or not. But why the training speed will slows down a lot sometimes? What should I do to improve the %util?

I sit in a Tesla and translated this thread with Ai:

mangohost

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.