From: Gerrit Renker Date: Fri, 03 Aug 2007 14:14:16 +0000 Subject: Re: relating code to RFC 4342, RFC 3448, rfc3448bis-00 Message-Id: <200708031514.16519@strip-the-willow> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable To: dccp@vger.kernel.org Hi Ian, below is as promised the list of differences of the CCID3 code, this has th= e most relevant differences. The list may not be exhaustive, but this is all I can get hold of for the m= oment. Most is also documented online at http://www.erg.abdn.ac.uk/users/gerrit/dccp/notes/ccid3_packet_recept= ion/ 1. General ---------- The code conforms to RFC 4342 as far as possible. Where deviating, the chan= ges introduced in draft rfc3448bis, revision 00, are used as reference. 2. TFRC code ------------- * loss detection is fully compliant with RFC 4342, including the handling o= f DUPACKS * details on=20 http://www.erg.abdn.ac.uk/users/gerrit/dccp/notes/ccid3_packet_reception/= loss_detection/loss_detection_algorithm_notes.txt=20 * receiver RTT sampling uses a modified algorithm from RFC 4342, 8.1. This = is an extension (uses all data it can get), but the principle is the algorithm from RFC 4= 342. * the loss interval handling uses rfc3448bis rather than RFC 3448, due to t= he following limitation of RFC 3448:=20 RFC 3448 considers only a fixed set of n=3D8 loss intervals. It takes quit= e a while until that many loss intervals have accumulated. rfc3448bis is bett= er in this regard, since it specifies handling up to k < n =3D 8 loss intervals. This = is more realistic=20 and hence has been adapted for the code. =20 * the weights used for the individual loss intervals use the precise, not t= he mod-2 rounded values, from RFC 3448, section 5.4 rather than section 8. This is because= it gives a better performance and since there is no real gain in using shift operations her= e. * lost packets are always assumed to be data packets (cf. section 7 of the = above URL for clarification) * updating I_mean, in contrast, is based on section 10.2 of RFC 4342 rather= than RFC 3448/rfc3448bis. RFC 4342 introduced a method to find out where a new loss interval starts= , with respect to sequence number wrap-around (field li_is_closed). * similarly, determining whether a newly identified loss event does indeed = designate a new loss=20 interval depends is based on RFC 4342, 10.2 rather than RFC 3448/rfc3448b= is 3. CCID3 code -------------- * RFC3390 initial rate uses `s' instead of MSS described in RFC 4342, as pe= r previous discussion * all non-RFC 4342 based code uses revision 00 of draft rfc3448bis as basis= , in particular: - computation procedures in ccid3_hc_tx_update_x(), - updating X when feedback is received in ccid3_hc_tx_packet_recv(), - updating X in ccid3_hc_tx_no_feedback_timer(), * TX CCID3 uses RTT from Request/Response handshake - as per RFC 4342 errat= um. * packet scheduling closely follows section 4.6 of RFC 3448. Still not sure= that this is the best way to do this, but recent tests (cf. bug reports and dropping of `send c= redit' patch) showed that apparently this works quite satisfactorily. * the value of `s' is not fixed but uses EWMA - something which rfc3448bis = clarified and introduced. * computing X_recv uses a disambiguation which tries to reconcile between t= he different definitions of RFC 3448 and 4342, cf.=20 http://www.erg.abdn.ac.uk/users/gerrit/dccp/notes/ccid3_packet_reception/= #8._Computing_X_recv_ * introduced the possibility of changing the t_nfb to avoid frequently call= ing the no_feedback_timer when the RTT is small (Kconfig option) * CCID3 packet reception (RX CCID) is largely conformant with RFC 3448/rfc3= 448bis rev 00 and was a pain to get right. See http://www.erg.abdn.ac.uk/users/gerrit/dccp/notes/= ccid3_packet_reception why. * sending feedback is also largely conformant with RFC 4342/3448, with the = addition of the clarification for computing X_recv when the time interval between sending the last feed= back and now is less than one RTT - in this case the previously computed X_recv value is reused (would = otherwise lead to false=20 estimates of X_recv). HTH Gerrit | > | =A0If you do have a list of them it would be great. I see quite a few | > | =A0comments about bis in the code. | > I don't have a list at present, need to go through the works. Will pos= t at a later point. | > | Don't worry too much about this, as this is my research so don't want | to waste your time :-)