Veek M
Disciple
https://www.kernel.org/doc/html/lat...erview.html#block-and-inode-allocation-policy
Basically says that the FS tries to keep things sequential in 128MB chunks and spread the 128MB chunks out across the entire disk at the directory level while keeping the inode close to the dir.
Which means that the SSD firmware is going to see sequential-block-number queries/commands for files larger than 4k up-to 128MB.
Why then does the 4k value matter - well because I just averaged my FS files and it's 6k-40k for anything less than 100k depending on what directories you chose to average. So for OS load the 4k is more relevant. Even the Firefox dir is 3MB avg.
However if you read that Linux article, it's also storing based on dir contents so you can expect sequential access based on directory locality (up-to 128MB). Which puts it in the realm of sequential access - basically.
(so that 4k thing is no longer that relevant)
Right|Wrong - why?
(excluding the fact that a good 4kIOPS disk is likely to be better manufactured - intangibles and such)
Basically says that the FS tries to keep things sequential in 128MB chunks and spread the 128MB chunks out across the entire disk at the directory level while keeping the inode close to the dir.
Which means that the SSD firmware is going to see sequential-block-number queries/commands for files larger than 4k up-to 128MB.
Why then does the 4k value matter - well because I just averaged my FS files and it's 6k-40k for anything less than 100k depending on what directories you chose to average. So for OS load the 4k is more relevant. Even the Firefox dir is 3MB avg.
However if you read that Linux article, it's also storing based on dir contents so you can expect sequential access based on directory locality (up-to 128MB). Which puts it in the realm of sequential access - basically.
(so that 4k thing is no longer that relevant)
Right|Wrong - why?
(excluding the fact that a good 4kIOPS disk is likely to be better manufactured - intangibles and such)
