PATCH: killing read_ahead[]
Quoted 4 lines · Linus Torvalds
Actually, the _real_ answer is to make
fs/block_dev.cuse the page cache instead - andgeneric_file_read()does read-ahead that actually improves performance, unlike the silly contortions that the direct block-dev read-ahead tries to do.
If we had a paper about the page cache this would be easy.
In the beginning page cache was just previously mmaped pages, that are clean and ready to be mapped again.
Today we have them either dirty or clean, mapped or not(?), with and without buffers, in highmem(?) or lowmem and everybody and its children is using it for everything.
We need a clear definition about (concurrent) states of page cached pages, valid transitions (and locks/sema4s to take for them), assumptions, guarantees etc.
The only thing I see guaranteed, that every big thing to be cached should live there and is page aligned and page sized.
I'm trying hard to understand a concept in the page cache and to get it's limits and guarantees, but still find it hard to get them.
Time for specs, I would say ;-)
I could help to explain and formulate, if someone could only cut the edges of how it works and what it will be.
Thanks & Regards
Ingo Oeser
Signature · 6 lines
Feel the power of the penguin - run [email protected] <esc>:x -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM, see: http://www.linux.eu.org/Linux-MM/
Quoted 26 lines · Ingo Oeser
Quoted 4 lines · Linus Torvalds
Actually, the _real_ answer is to make
fs/block_dev.cuse the page cache instead - andgeneric_file_read()does read-ahead that actually improves performance, unlike the silly contortions that the direct block-dev read-ahead tries to do.If we had a paper about the page cache this would be easy.
In the beginning page cache was just previously mmaped pages, that are clean and ready to be mapped again.
Today we have them either dirty or clean, mapped or not(?), with and without buffers, in highmem(?) or lowmem and everybody and its children is using it for everything.
We need a clear definition about (concurrent) states of page cached pages, valid transitions (and locks/sema4s to take for them), assumptions, guarantees etc.
The only thing I see guaranteed, that every big thing to be cached should live there and is page aligned and page sized.
I'm trying hard to understand a concept in the page cache and to get it's limits and guarantees, but still find it hard to get them.
Quoted 12 lines
Time for specs, I would say ;-)
I could help to explain and formulate, if someone could only cut the edges of how it works and what it will be.
Thanks & Regards
Ingo Oeser
Signature · 3 lines
Feel the power of the penguin - run [email protected] <esc>:x -
I hope we are not doing something stupid here, like breaking the f*!%cking page cache again. I've finaly got all the bugs out of NWFS on 2.4.0-test9, and have waded through the breakage of the past two testX releases of 2.4.
Why do we need to disable read ahead on the page cache anyway?
:-)
Jeff
Quoted 3 lines
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/