On Mon, Mar 17, 2003 at 12:55:17AM +0300, Alex Tomas wrote:
> > Also, while your goal is to reduce cache ping-pong between CPUs, we will
> > now have cache ping-pong for the "diff" array. We need to do per-cpu
> > values or make each value cacheline aligned to avoid ping-pong.
>
> yes, you're right. fixed
Um. But ... let's say I'm on a P4 box running a distro kernel. I suspect
most distros will set NR_CPUs to 8. So that's 128 byte cachelines * 8 CPUs
gives 1024 bytes -- 1k. Per dcounter (never mind the size of seqlock or
dc_base/dc_min):
> +struct dcounter_diff {
> + long dd_value;
> +} ____cacheline_aligned_in_smp;
> +
> +struct dcounter {
> + long dc_base;
> + long dc_min;
> + struct dcounter_diff dc_diff[NR_CPUS];
> + seqlock_t dc_lock;
> +};
> +++ edited/include/linux/ext2_fs_sb.h Mon Mar 17 00:12:00 2003
> struct ext2_bg_info *s_bgi;
> + struct dcounter free_blocks_dc;
> + struct dcounter free_inodes_dc;
> + struct dcounter dirs_dc;
And then we have 3 of these (an additional 3k..). Per blockgroup.
My 4GB / has 30 blockgroups; my 30GB /home has 232. So that's a little
under 8 per GB. My _laptop_ has a 40GB drive, so that's on the order
of 320 blockgroups -- almost an additional megabyte of ram consumed for
these counters.
Maybe it makes big boxes go faster, but this makes my dual CPU desktop
go slower, and that's not acceptable.
-- "It's not Hollywood. War is real, war is primarily not about defeat or victory, it is about death. I've seen thousands and thousands of dead bodies. Do you think I want to have an academic debate on this subject?" -- Robert Fisk - 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/
This archive was generated by hypermail 2b29 : Sun Mar 23 2003 - 22:00:20 EST