Return-path: Received: from mail.gmx.net ([213.165.64.20]:52378 "HELO mail.gmx.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with SMTP id S1753468AbXLIW3Z (ORCPT ); Sun, 9 Dec 2007 17:29:25 -0500 Subject: Re: [RFC/T][PATCH 2/3] rc80211-pid: introduce PID sharpening factor From: Mattias Nissler To: Stefano Brivio Cc: linux-wireless , "John W. Linville" , Johannes Berg In-Reply-To: <20071209212133.43030d03@morte> References: <20071209211547.2d7fca32@morte> <20071209212133.43030d03@morte> Content-Type: text/plain Date: Sun, 09 Dec 2007 23:29:21 +0100 Message-Id: <1197239361.7543.17.camel@localhost> (sfid-20071209_222950_791474_5B899962) Mime-Version: 1.0 Sender: linux-wireless-owner@vger.kernel.org List-ID: On Sun, 2007-12-09 at 21:21 +0100, Stefano Brivio wrote: > This patch introduces a PID sharpening factor for faster response on > association and interpolation. > > > Signed-off-by: Stefano Brivio > > --- > > Index: wireless-2.6/net/mac80211/rc80211_pid.c > =================================================================== > --- wireless-2.6.orig/net/mac80211/rc80211_pid.c > +++ wireless-2.6/net/mac80211/rc80211_pid.c > @@ -23,13 +23,15 @@ > * > * The controller basically computes the following: > * > - * adj = CP * err + CI * err_avg + CD * (err - last_err) > + * adj = CP * err + CI * err_avg + CD * (err - last_err) * (1 + sharpening) > * > * where > * adj adjustment value that is used to switch TX rate (see below) > * err current error: target vs. current failed frames percentage > * last_err last error > * err_avg average (i.e. poor man's integral) of recent errors > + * sharpening non-zero when fast response is needed (i.e. right after > + * association or interpolation), heading to zero over time > * CP Proportional coefficient > * CI Integral coefficient > * CD Derivative coefficient > @@ -65,6 +67,11 @@ > #define RC_PID_SMOOTHING_SHIFT 3 > #define RC_PID_SMOOTHING (1 << RC_PID_SMOOTHING_SHIFT) > > +/* Sharpening factor (used for D part of PID controller) */ > +#define RATE_CONTROL_SHARPENING_SHIFT 2 > +#define RATE_CONTROL_SHARPENING (1 << RATE_CONTROL_SHARPENING_SHIFT) > +#define RATE_CONTROL_SHARPENING_DURATION 1 > + > /* Fixed point arithmetic shifting amount. */ > #define RC_PID_ARITH_SHIFT 8 > > @@ -127,8 +134,11 @@ struct rc_pid_sta_info { > */ > s32 err_avg_sc; > > - /* Last framed failes percentage sample */ > + /* Last framed failes percentage sample. */ > u32 last_pf; > + > + /* Sharpening needed. */ > + u8 sharp_cnt; > }; > > /* Algorithm parameters. We keep them on a per-algorithm approach, so they can > @@ -271,10 +281,12 @@ static void rate_control_pid_sample(stru > spinfo->last_sample = jiffies; > > /* If no frames were transmitted, we assume the old sample is > - * still a good measurement and copy it. */ > - if (spinfo->tx_num_xmit == 0) > + * still a good measurement and copy it, and turn the sharpening factor > + * on. */ > + if (spinfo->tx_num_xmit == 0) { > pf = spinfo->last_pf; > - else { > + spinfo->sharp_cnt = RATE_CONTROL_SHARPENING_DURATION; > + } else { Note that current rate_control_pid_sample() is only called from rate_control_pid_tx_status(), which does an tx_num_xmit++ in advance. So the tx_num_xmit branch should actually never be executed (I kept it only to guard against any division by zero errors). I guess this makes the whole patch basically a noop. > pf = spinfo->tx_num_failed * 100 / spinfo->tx_num_xmit; > pf <<= RC_PID_ARITH_SHIFT; > > @@ -314,8 +326,11 @@ static void rate_control_pid_sample(stru > spinfo->err_avg_sc = spinfo->err_avg_sc - err_avg + err_prop; > err_int = spinfo->err_avg_sc >> RC_PID_SMOOTHING_SHIFT; > > - err_der = pf - spinfo->last_pf; > + err_der = pf - spinfo->last_pf > + * (1 + RATE_CONTROL_SHARPENING * spinfo->sharp_cnt); > spinfo->last_pf = pf; > + if (spinfo->sharp_cnt) > + spinfo->sharp_cnt--; > > /* Compute the controller output. */ > adj = (err_prop * pinfo->coeff_p + err_int * pinfo->coeff_i > >