[rrd-developers] Implementation of temporal smoothing

[ch_cu2 at yahoo.fr]

Having a look at the source code, I can't really understand the logic of 
the implementation of the temporal smoothing in rrdtool. I re-read the 
spec and have no problem with them, but the implementation remains 
unclear for me.

I put a copy of the code in question below.

Could someone explain me what the code is doing and why we don't simply 
do a temp smoothing at, say, every burning cycle at a fixed row (which 
would be fixed for each rrd file and generated randomly while it is 
created). This would be much simpler... but I must miss something.

Also, why is the BURNIN_CYCLES constant set to 3?

Thanks for your help!

cheers,
Chris


Code:
                    /* periodically run a smoother for seasonal effects */
                    /* Need to use first cdp parameter buffer to track 
burnin (burnin requires a specific smoothing schedule).
                    * The CDP_init_seasonal parameter is really an RRA 
level, not a data source within RRA level parameter, but the
                    * rra_def is read only for rrd_update (not flushed 
to disk). */
                    iii = i*(rrd.stat_head -> ds_cnt);
                    if 
(rrd.cdp_prep[iii].scratch[CDP_init_seasonal].u_cnt  <= BURNIN_CYCLES)
                    {
                        if (rrd.rra_ptr[i].cur_row + elapsed_pdp_st > 
rrd.rra_def[i].row_cnt - 1)
                        {
                            /* mark off one of the burnin cycles */
                            
++(rrd.cdp_prep[iii].scratch[CDP_init_seasonal].u_cnt);
                            schedule_smooth = 1;
                        } 
                    }
                    else
                    {
                        /* someone has no doubt invented a trick to deal 
with this wrap around, but at least this code is clear. */
                        if 
(rrd.rra_def[i].par[RRA_seasonal_smooth_idx].u_cnt > rrd.rra_ptr[i].cur_row)
                        {
                            /* here elapsed_pdp_st = rra_step_cnt[i] 
because of 1-1 mapping between PDP and CDP */
                            if (rrd.rra_ptr[i].cur_row + elapsed_pdp_st 
 >= rrd.rra_def[i].par[RRA_seasonal_smooth_idx].u_cnt)
                            {
                                #ifdef DEBUG
                                    fprintf(stderr,
                                        "schedule_smooth 1: cur_row %lu, 
elapsed_pdp_st %lu, smooth idx %lu\n",
                                        rrd.rra_ptr[i].cur_row, 
elapsed_pdp_st,
                                        
rrd.rra_def[i].par[RRA_seasonal_smooth_idx].u_cnt);
                                #endif
                                schedule_smooth = 1;
                            }
                        }
                        else
                        {
                            /* can't rely on negative numbers because we 
are working with unsigned values */
                            /* Don't need modulus here. If we've wrapped 
more than once, only one smooth is executed at the end. */
                            if (rrd.rra_ptr[i].cur_row + elapsed_pdp_st 
 >= rrd.rra_def[i].row_cnt
                                && rrd.rra_ptr[i].cur_row + 
elapsed_pdp_st - rrd.rra_def[i].row_cnt
                                >= 
rrd.rra_def[i].par[RRA_seasonal_smooth_idx].u_cnt)
                            {
                                #ifdef DEBUG
                                    fprintf(stderr,
                                        "schedule_smooth 2: cur_row %lu, 
elapsed_pdp_st %lu, smooth idx %lu\n",
                                        rrd.rra_ptr[i].cur_row, 
elapsed_pdp_st,
                                        
rrd.rra_def[i].par[RRA_seasonal_smooth_idx].u_cnt);
                                #endif
                                schedule_smooth = 1;
                            }
                        }
                    }






	
 p5.vert.ukl.yahoo.com uncompressed Fri Sep  1 13:27:00 GMT 2006 
	
		
___________________________________________________________________________ 
Yahoo! Mail réinvente le mail ! Découvrez le nouveau Yahoo! Mail et son interface révolutionnaire.
http://fr.mail.yahoo.com

--
Unsubscribe mailto:rrd-developers-request at list.ee.ethz.ch?subject=unsubscribe
Help        mailto:rrd-developers-request at list.ee.ethz.ch?subject=help
Archive     http://lists.ee.ethz.ch/rrd-developers
WebAdmin    http://lists.ee.ethz.ch/lsg2.cgi