I've done substantial AIM7 benchmarking on a 28p ia64 NUMA system, and
llseek's BKL usage is a significant contributor to poor scaling. For
500 AIM7 "tasks" and ext2 filesystems, waiting on the BKL consumes about
half of the available CPU cycles, and sys_lseek()'s usage is the most
significant cycle waster, followed by ext2_get_block() and
ext2_write_inode(). Anton's llseek patch from last November does make
a measurable improvement in AIM7 throughput.
-- John Hawkes hawkes@sgi.com
- 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/