topics I'd like to discuss
A message that is not in the archive (sent off-list or not yet received): [email protected]
Quoted 3 lines · Larry Woodman
1.) Using mmu notifiers to set up the page tables for integrated devices(GPUs) and allowing the generic
kernel pagefault handler to resolve translations for those devices.
This functionality is also desired by people who want to run memory coherency over infiniband and some other very fast network fabrics, as well as by the people who want to offload work to specialized cores in their system.
Since there are multiple use cases for this kind of functionality, I believe that it is important we get the infrastructure right, and discussing this topic at LSF/MM sounds like a good idea.
Quoted 1 line
2.) Replication of pagecache pages on NUMA nodes.
What about this would you like to discuss?
Is there some proposal of code to do this?
Signature · 7 lines
All rights reversed -- 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-mm.org/ . Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Quoted 6 lines · Rik van Riel
Quoted 4 lines · Larry Woodman
1.) Using mmu notifiers to set up the page tables for integrated devices(GPUs) and allowing the generic
kernel pagefault handler to resolve translations for those
devices.
Quoted 8 lines
This functionality is also desired by people who want to run memory coherency over infiniband and some other very fast network fabrics, as well as by the people who want to offload work to specialized cores in their system.
Since there are multiple use cases for this kind of functionality, I believe that it is important we get the infrastructure right, and discussing this topic at LSF/MM sounds like a good idea.
Agreed.
Quoted 4 lines
Quoted 2 lines
2.) Replication of pagecache pages on NUMA nodes.
What about this would you like to discuss?
The tradeoffs between local memory access for frequently referenced read-only/executable pages like shared library text and the additional memory consumption associated replicating pages on multiple NUMA nodes.
Quoted 2 lines
Is there some proposal of code to do this?
Not yet, I implemented replication of filesystem cache memory on each NUMA node several years ago for a UNIX kernel when NUMA systems first appeared. At that time both memory and caches were much smaller than they are now so the tradeoffs between local memory access and inducing page reclamation were different than they are now. However there was a significant performance boost on memory rich systems running applications with bloated text sections.
Larry
--
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-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Quoted 5 lines · Rik van Riel
Quoted 1 line
2.) Replication of pagecache pages on NUMA nodes.
What about this would you like to discuss?
Is there some proposal of code to do this?
I did an implementation _long_ ago:
http://lwn.net/Articles/63512/
Might be fun for someone to glean an idea or two from, although I'm sure the code has long since bitrotted in to being useless. It was kinda fun to read, though. :)
--
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-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>