Re: [Patch 5/6] statistics infrastructure

[Nikita Danilov <nikita@clusterfs.com>]

Martin Peschke writes:
 > This patch adds statistics infrastructure as common code.
 > 

[...]

 > +
 > +static void statistic_add_util(struct statistic *stat, int cpu,
 > +			       s64 value, u64 incr)
 > +{
 > +	struct statistic_entry_util *util = stat->pdata->ptrs[cpu];
 > +	util->num += incr;
 > +	util->acc += value * incr;
 > +	if (unlikely(value < util->min))
 > +		util->min = value;
 > +	if (unlikely(value > util->max))
 > +		util->max = value;

One useful aggregate that can be calculated here is a standard
deviation. Something like

        util->acc_sq += value * incr * value * incr; /* sum of squares */

And in statistic_fdata_util() squared standard deviation is

        std_dev = util->acc_sq;
        /*
         * Difference of averaged square and squared average.
         */
        std_dev = do_div(std_dev, util->num) - whole * whole;

 > +}

Nikita.