This patch series helps to resolve scalability problem
for tmpfs.  With these patches, Aim7 fserver throughput for tmpfs
improved by 270% on a 4 socket, 32 cores NHM-EX system.

In current implementation of tmpfs, whenever we
get a new page, stat_lock in shmem_sb_info needs to be acquired.  
This causes a lot of lock contentions when multiple
threads are using tmpfs simultaneously, which makes
system with large number of cpus scale poorly.
Almost 75% of cpu time was spent contending on
stat_lock when we ran Aim7 fserver load with 128 threads
on a 4 socket, 32 cores NHM-EX system.

We made use of the percpu_counter library for used blocks accounting to
speed up the getting and returning of blocks to local per cpu counter
without lock acquisition.

The first patch in the series add a function to provide
fast but accurate comparison for the percpu_counter library. 

The second patch update the shmem code of tmpfs to use percpu_counter 
library to improve tmpfs performance.

Version 3 Changes:
1. Use percpu_counter instead of adding a new qtoken library for fast
block accounting.
2. Change accounting from free_blocks to used_blocks so we do not need
to reset percpu_counter when we remount tmpfs.


Regards,
Tim Chen




--
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/