| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Tue, 29 Jul 2008 11:05:50 +0100
From: Gerrit Renker <gerrit@....abdn.ac.uk>
To: dccp@...r.kernel.org
Cc: netdev@...r.kernel.org, Gerrit Renker <gerrit@....abdn.ac.uk>
Subject: [PATCH 5/7] dccp tfrc: Increase number of RTT samples
This improves the receiver RTT sampling algorithm so that it tries harder to get
as many RTT samples as possible, using several optimisations.
Applicability and background
----------------------------
The algorithm uses timestamps and differences of the CCval window counter to
guess the RTT, using concepts presented in RFC 4340, 8.1.
There exist 4 cases for the CCVal difference:
* == 0: less than RTT/4 passed since last packet -- unusable sample;
* > 4: (much) more than 1 RTT has passed since last packet -- also unusable;
* == 4: "perfect" sample (exactly one RTT has passed since last packet);
* 1..3: sub-optimal sample (between RTT/4 and 3*RTT/4 has passed).
In the last case the algorithm so far tried to optimise by storing away the
candidate and then re-trying next time. This had the following problems:
* a large number of samples is needed to smooth out the given inaccuracies;
* the sender may not be sending enough packets to warrant a "next time";
* hence it is better to use suboptimal samples whenever possible.
As a consequence, the algorithm now stores away the current sample only if the
difference is 0.
A realistic example to illustrate the failure of the (previous) algorithm is MP3
streaming, where packets are sent at a rate of less than one packet per RTT.
Which means that suitable samples may be absent for a very long time.
The effectiveness of using suboptimal samples (with a delta between 1 and 4) was
confirmed by instrumenting the algorithm with counters. The results of two 20
second test runs were:
* With the old algorithm and a total of 38442 function calls, only 394 of these
calls resulted in usable RTT samples (about 1%), 378 out of these were
"perfect" samples, and 28013 (unused) samples had a delta of 1..3.
* With the new algorithm and a total of 37057 function calls, 1702 usable RTT
samples were retrieved (about 4.6%), 5 out of these were "perfect" samples.
This means an almost five-fold increase in the number of samples.
Signed-off-by: Gerrit Renker <gerrit@....abdn.ac.uk>
---
net/dccp/ccids/lib/packet_history.c | 83 ++++++++++-------------------------
1 files changed, 24 insertions(+), 59 deletions(-)
--- a/net/dccp/ccids/lib/packet_history.c
+++ b/net/dccp/ccids/lib/packet_history.c
@@ -428,31 +428,16 @@ int tfrc_rx_hist_init(struct tfrc_rx_hist *h, struct sock *sk)
EXPORT_SYMBOL_GPL(tfrc_rx_hist_init);
/**
- * tfrc_rx_hist_rtt_last_s - reference entry to compute RTT samples against
- */
-static inline struct tfrc_rx_hist_entry *
- tfrc_rx_hist_rtt_last_s(const struct tfrc_rx_hist *h)
-{
- return h->ring[0];
-}
-
-/**
- * tfrc_rx_hist_rtt_prev_s: previously suitable (wrt rtt_last_s) RTT-sampling entry
- */
-static inline struct tfrc_rx_hist_entry *
- tfrc_rx_hist_rtt_prev_s(const struct tfrc_rx_hist *h)
-{
- return h->ring[h->rtt_sample_prev];
-}
-
-/**
* tfrc_rx_hist_sample_rtt - Sample RTT from timestamp / CCVal
- * Based on ideas presented in RFC 4342, 8.1. Returns 0 if it was not able
- * to compute a sample with given data - calling function should check this.
+ * Based on ideas presented in RFC 4342, 8.1. This function expects that no loss
+ * is pending and uses the following history entries (via rtt_sample_prev):
+ * - h->ring[0] contains the most recent history entry prior to @skb;
+ * - h->ring[1] is an unused `dummy' entry when the current difference is 0;
*/
void tfrc_rx_hist_sample_rtt(struct tfrc_rx_hist *h, const struct sk_buff *skb)
{
- u32 sample = 0, delta_v;
+ struct tfrc_rx_hist_entry *last = h->ring[0];
+ u32 sample, delta_v;
/*
* When not to sample:
@@ -466,47 +451,27 @@ void tfrc_rx_hist_sample_rtt(struct tfrc_rx_hist *h, const struct sk_buff *skb)
tfrc_rx_hist_loss_pending(h))
return;
- delta_v = SUB16(dccp_hdr(skb)->dccph_ccval,
- tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
-
- if (delta_v < 1 || delta_v > 4) { /* unsuitable CCVal delta */
- if (h->rtt_sample_prev == 2) { /* previous candidate stored */
- sample = SUB16(tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_ccval,
- tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
- if (sample)
- sample = 4 / sample *
- ktime_us_delta(tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_tstamp,
- tfrc_rx_hist_rtt_last_s(h)->tfrchrx_tstamp);
- else /*
- * FIXME: This condition is in principle not
- * possible but occurs when CCID is used for
- * two-way data traffic. I have tried to trace
- * it, but the cause does not seem to be here.
- */
- DCCP_BUG("please report to dccp@...r.kernel.org"
- " => prev = %u, last = %u",
- tfrc_rx_hist_rtt_prev_s(h)->tfrchrx_ccval,
- tfrc_rx_hist_rtt_last_s(h)->tfrchrx_ccval);
- } else if (delta_v < 1) {
- h->rtt_sample_prev = 1;
- goto keep_ref_for_next_time;
- }
-
- } else if (delta_v == 4) /* optimal match */
- sample = ktime_to_us(net_timedelta(tfrc_rx_hist_rtt_last_s(h)->tfrchrx_tstamp));
- else { /* suboptimal match */
- h->rtt_sample_prev = 2;
- goto keep_ref_for_next_time;
- }
+ h->rtt_sample_prev = 0; /* reset previous candidate */
- if (unlikely(sample > DCCP_SANE_RTT_MAX)) {
- DCCP_WARN("RTT sample %u too large, using max\n", sample);
- sample = DCCP_SANE_RTT_MAX;
+ delta_v = SUB16(dccp_hdr(skb)->dccph_ccval, last->tfrchrx_ccval);
+ if (delta_v == 0) { /* less than RTT/4 difference */
+ h->rtt_sample_prev = 1;
+ return;
}
+ sample = dccp_sane_rtt(ktime_to_us(net_timedelta(last->tfrchrx_tstamp)));
- h->rtt_sample_prev = 0; /* use current entry as next reference */
-keep_ref_for_next_time:
+ if (delta_v <= 4) /* between RTT/4 and RTT */
+ sample *= 4 / delta_v;
+ else if (!(sample < h->rtt_estimate && sample > h->rtt_estimate/2))
+ /*
+ * Optimisation: CCVal difference is greater than 1 RTT, yet the
+ * sample is less than the local RTT estimate; which means that
+ * the RTT estimate is too high.
+ * To avoid noise, it is not done if the sample is below RTT/2.
+ */
+ return;
- h->rtt_estimate = tfrc_ewma(h->rtt_estimate, sample, 9);
+ /* Use a lower weight than usual to increase responsiveness */
+ h->rtt_estimate = tfrc_ewma(h->rtt_estimate, sample, 5);
}
EXPORT_SYMBOL_GPL(tfrc_rx_hist_sample_rtt);
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists