: Disable broken Ack Ratio adaptation algorithm
PATCH 1/9 : Disable broken Ack Ratio adaptation algorithm
This comments out a problematic section comprising a half-finished algorithm:
- The variable `ccid2hctx_ackloss' is never initialised to a value different from 0 and hence in fact is a read-only constant.
- The `arsent' variable counts packets other than Acks (it is incremented for every packet), and there is no test for Ack Loss.
- The concept of counting Acks as such leads to a complex calculation, and the calculation at the moment is inconsistent with this concept.
The problem is that the number of Acks - rather than the number of windows - is counted, which leads to a complex (cubic/quadratic) expression - this is not even implemented.
In its current state, the commented-out algorithm interferes with normal processing by changing Ack Ratio incorrectly, and at the wrong times.
A new algorithm is necessary, which will not necessarily use the same variables as used by the unfinished one; hence the old variables have been removed.
--- net/dccp/ccids/ccid2.c | 23 +++++++++++++++++++++-- net/dccp/ccids/ccid2.h | 4 ---- 2 files changed, 21 insertions(+), 6 deletions(-)
net/dccp/ccids/ccid2.c +21 -2
@@ -224,8 +224,6 @@ static void ccid2_hc_tx_rto_expire(unsigned long data) hctx->ccid2hctx_sent =0 ;/* clear ack ratio state. */ - hctx->ccid2hctx_arsent =0 ; - hctx->ccid2hctx_ackloss =0 ; hctx->ccid2hctx_rpseq =0 ; hctx->ccid2hctx_rpdupack =-1 ;ccid2_change_l_ack_ratio (sk,1 );
@@ -289,6 +287,26 @@ static void ccid2_hc_tx_packet_sent(struct sock *sk, int more, unsigned int len) hctx->ccid2hctx_sent++; +/* +* FIXME: The code below is broken and the variables have been removed +* from the socket struct. The `ackloss' variable was always set to 0, +* and with arsent there are several problems: +* (i) it doesn't just count the number of Acks, but all sent packets; +* (ii) it is expressed in # of packets, not # of windows, so the +* comparison below uses the wrong formula: Appendix A of RFC 4341 +* comes up with the number K = cwnd / (R^2 - R) of consecutive windows +* of data with no lost or marked Ack packets. If arsent were the # of +* consecutive Acks received without loss, then Ack Ratio needs to be +* decreased by 1 when +* arsent >= K * cwnd / R = cwnd^2 / (R^3 - R^2) +* where cwnd / R is the number of Acks received per window of data +* (cf. RFC 4341, App. A). The problems are that +* - arsent counts other packets as well; +* - the comparison uses a formula different from RFC 4341; +* - computing a cubic/quadratic equation each time is too complicated. +* Hence a different algorithm is needed. +*/ +#if 0 /* Ack Ratio. Need to maintain a concept of how many windows we sent */ hctx->ccid2hctx_arsent++;/* We had an ack loss in this window... */
@@ -316,6 +334,7 @@ static void ccid2_hc_tx_packet_sent(struct sock *sk, int more, unsigned int len) hctx->ccid2hctx_arsent = 0; /* or maybe set it to cwnd*/ } } +#endif /* setup RTO timer */ if (!timer_pending (&hctx->ccid2hctx_rtotimer))
net/dccp/ccids/ccid2.h +0 -4
@@ -44,8 +44,6 @@ struct ccid2_seq {* @ccid2hctx_acks - ACKS recv in AI phase * @ccid2hctx_sent - packets sent in this window * @ccid2hctx_lastrtt -time RTT was last measured -* @ccid2hctx_arsent - packets sent [ack ratio] -* @ccid2hctx_ackloss - ack was lost in this win * @ccid2hctx_rpseq - last consecutive seqno * @ccid2hctx_rpdupack - dupacks since rpseq */
@@ -66,8 +64,6 @@ struct ccid2_hc_tx_sock {int ccid2hctx_sent;unsigned long ccid2hctx_lastrtt;struct timer_list ccid2hctx_rtotimer; -unsigned long ccid2hctx_arsent; -int ccid2hctx_ackloss;u64 ccid2hctx_rpseq;int ccid2hctx_rpdupack;int ccid2hctx_sendwait;
Signature · 1 line
1.5.2.2.238.g7cbf2f2-dirty
Quoted 11 lines · Gerrit Renker
This comments out a problematic section comprising a half-finished algorithm:
- The variable `ccid2hctx_ackloss' is never initialised to a value different from 0 and hence in fact is a read-only constant.
- The `arsent' variable counts packets other than Acks (it is incremented for every packet), and there is no test for Ack Loss.
- The concept of counting Acks as such leads to a complex calculation, and the calculation at the moment is inconsistent with this concept.
The problem is that the number of Acks - rather than the number of windows - is counted, which leads to a complex (cubic/quadratic) expression - this is not even implemented.
Quoted 7 lines
In its current state, the commented-out algorithm interferes with normal processing by changing Ack Ratio incorrectly, and at the wrong times.
A new algorithm is necessary, which will not necessarily use the same variables as used by the unfinished one; hence the old variables have been removed.
Signed-off-by Gerrit Renker <[email protected]>