Status update : Brick Mux threads reduction

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi,

For each brick, we create atleast 20+ threads, hence in a brick mux use case, where we load multiple bricks in the same process, there will 100s of threads resulting in perf issues, memory usage increase.

IO-threads :  Make it global, to the process, and ref count the resource. patch [1], has failures in brick mux regression, likey not related to the patch, need to get it passed.

Posix- threads : Janitor, Helper, Fsyncer, instead of using one thread per task, use synctask framework instead. In the future use thread pool in patch [2]. Patches are posted[1], fixing some regression failures.

Posix, bitrot aio-thread : This thread cannot be replaced to just use synctask/thread pool as there cannot be a delay in recieving notifications and acting on it. Hence, create a global aio event receiver thread for the process. This is WIP and is not yet posted upstream.

Threads in changelog/bitrot xlator Mohit posted a patch where default xlator does not need to start a thread if xlator is not enabled
https://review.gluster.org/#/c/glusterfs/+/21304/ (it can save 6 thread per brick in default option)

Pending: Create a build of these patches, run perf tests with these patches and analyze the same.


[1] https://review.gluster.org/#/c/glusterfs/+/20761/
[2] https://review.gluster.org/#/c/glusterfs/+/20636/

Regards,
Poornima
_______________________________________________
Gluster-devel mailing list
Gluster-devel@xxxxxxxxxxx
https://lists.gluster.org/mailman/listinfo/gluster-devel

[Index of Archives]     [Gluster Users]     [Ceph Users]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Security]     [Bugtraq]     [Linux]     [Linux OMAP]     [Linux MIPS]     [eCos]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux