linux-pinenote/net/dccp/ccids/lib/packet_history.c

507 lines
15 KiB
C
Raw Normal View History

/*
* net/dccp/packet_history.c
*
* Copyright (c) 2007 The University of Aberdeen, Scotland, UK
* Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
*
* An implementation of the DCCP protocol
*
* This code has been developed by the University of Waikato WAND
* research group. For further information please see http://www.wand.net.nz/
* or e-mail Ian McDonald - ian.mcdonald@jandi.co.nz
*
* This code also uses code from Lulea University, rereleased as GPL by its
* authors:
* Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
*
* Changes to meet Linux coding standards, to make it meet latest ccid3 draft
* and to make it work as a loadable module in the DCCP stack written by
* Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
*
* Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
*/
#include <linux/string.h>
#include <linux/slab.h>
#include "packet_history.h"
#include "../../dccp.h"
/*
* Transmitter History Routines
*/
static struct kmem_cache *tfrc_tx_hist_slab;
int __init tfrc_tx_packet_history_init(void)
{
tfrc_tx_hist_slab = kmem_cache_create("tfrc_tx_hist",
sizeof(struct tfrc_tx_hist_entry),
0, SLAB_HWCACHE_ALIGN, NULL);
return tfrc_tx_hist_slab == NULL ? -ENOBUFS : 0;
}
void tfrc_tx_packet_history_exit(void)
{
if (tfrc_tx_hist_slab != NULL) {
kmem_cache_destroy(tfrc_tx_hist_slab);
tfrc_tx_hist_slab = NULL;
}
}
int tfrc_tx_hist_add(struct tfrc_tx_hist_entry **headp, u64 seqno)
{
struct tfrc_tx_hist_entry *entry = kmem_cache_alloc(tfrc_tx_hist_slab, gfp_any());
if (entry == NULL)
return -ENOBUFS;
entry->seqno = seqno;
entry->stamp = ktime_get_real();
entry->next = *headp;
*headp = entry;
return 0;
}
EXPORT_SYMBOL_GPL(tfrc_tx_hist_add);
void tfrc_tx_hist_purge(struct tfrc_tx_hist_entry **headp)
{
struct tfrc_tx_hist_entry *head = *headp;
while (head != NULL) {
struct tfrc_tx_hist_entry *next = head->next;
kmem_cache_free(tfrc_tx_hist_slab, head);
head = next;
}
*headp = NULL;
}
EXPORT_SYMBOL_GPL(tfrc_tx_hist_purge);
/*
* Receiver History Routines
*/
static struct kmem_cache *tfrc_rx_hist_slab;
int __init tfrc_rx_packet_history_init(void)
{
tfrc_rx_hist_slab = kmem_cache_create("tfrc_rxh_cache",
sizeof(struct tfrc_rx_hist_entry),
0, SLAB_HWCACHE_ALIGN, NULL);
return tfrc_rx_hist_slab == NULL ? -ENOBUFS : 0;
}
void tfrc_rx_packet_history_exit(void)
{
if (tfrc_rx_hist_slab != NULL) {
kmem_cache_destroy(tfrc_rx_hist_slab);
tfrc_rx_hist_slab = NULL;
}
}
static inline void tfrc_rx_hist_entry_from_skb(struct tfrc_rx_hist_entry *entry,
const struct sk_buff *skb,
const u64 ndp)
{
const struct dccp_hdr *dh = dccp_hdr(skb);
entry->tfrchrx_seqno = DCCP_SKB_CB(skb)->dccpd_seq;
entry->tfrchrx_ccval = dh->dccph_ccval;
entry->tfrchrx_type = dh->dccph_type;
entry->tfrchrx_ndp = ndp;
entry->tfrchrx_tstamp = ktime_get_real();
}
void tfrc_rx_hist_add_packet(struct tfrc_rx_hist *h,
const struct sk_buff *skb,
const u64 ndp)
{
struct tfrc_rx_hist_entry *entry = tfrc_rx_hist_last_rcv(h);
tfrc_rx_hist_entry_from_skb(entry, skb, ndp);
}
EXPORT_SYMBOL_GPL(tfrc_rx_hist_add_packet);
/* has the packet contained in skb been seen before? */
int tfrc_rx_hist_duplicate(struct tfrc_rx_hist *h, struct sk_buff *skb)
{
const u64 seq = DCCP_SKB_CB(skb)->dccpd_seq;
int i;
if (dccp_delta_seqno(tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno, seq) <= 0)
return 1;
for (i = 1; i <= h->loss_count; i++)
if (tfrc_rx_hist_entry(h, i)->tfrchrx_seqno == seq)
return 1;
return 0;
}
EXPORT_SYMBOL_GPL(tfrc_rx_hist_duplicate);
static void __tfrc_rx_hist_swap(struct tfrc_rx_hist *h, const u8 a, const u8 b)
{
struct tfrc_rx_hist_entry *tmp = h->ring[a];
h->ring[a] = h->ring[b];
h->ring[b] = tmp;
}
static void tfrc_rx_hist_swap(struct tfrc_rx_hist *h, const u8 a, const u8 b)
{
__tfrc_rx_hist_swap(h, tfrc_rx_hist_index(h, a),
tfrc_rx_hist_index(h, b));
}
/**
* tfrc_rx_hist_resume_rtt_sampling - Prepare RX history for RTT sampling
* This is called after loss detection has finished, when the history entry
* with the index of `loss_count' holds the highest-received sequence number.
* RTT sampling requires this information at ring[0] (tfrc_rx_hist_sample_rtt).
*/
static inline void tfrc_rx_hist_resume_rtt_sampling(struct tfrc_rx_hist *h)
{
__tfrc_rx_hist_swap(h, 0, tfrc_rx_hist_index(h, h->loss_count));
h->loss_count = h->loss_start = 0;
}
/*
* Private helper functions for loss detection.
*
* In the descriptions, `Si' refers to the sequence number of entry number i,
* whose NDP count is `Ni' (lower case is used for variables).
* Note: All __xxx_loss functions expect that a test against duplicates has been
* performed already: the seqno of the skb must not be less than the seqno
* of loss_prev; and it must not equal that of any valid history entry.
*/
static void __do_track_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u64 n1)
{
u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
s1 = DCCP_SKB_CB(skb)->dccpd_seq;
if (!dccp_loss_free(s0, s1, n1)) /* gap between S0 and S1 */
h->loss_count = 1;
}
static void __one_after_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u32 n2)
{
u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
s2 = DCCP_SKB_CB(skb)->dccpd_seq;
if (likely(dccp_delta_seqno(s1, s2) > 0)) { /* S1 < S2 */
h->loss_count = 2;
tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 2), skb, n2);
return;
}
/* S0 < S2 < S1 */
if (dccp_loss_free(s0, s2, n2)) {
u64 n1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_ndp;
if (dccp_loss_free(s2, s1, n1)) {
/* hole is filled: S0, S2, and S1 are consecutive */
tfrc_rx_hist_resume_rtt_sampling(h);
} else
/* gap between S2 and S1: just update loss_prev */
tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_loss_prev(h), skb, n2);
} else { /* gap between S0 and S2 */
/*
* Reorder history to insert S2 between S0 and S1
*/
tfrc_rx_hist_swap(h, 0, 3);
h->loss_start = tfrc_rx_hist_index(h, 3);
tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n2);
h->loss_count = 2;
}
}
/* return 1 if a new loss event has been identified */
static int __two_after_loss(struct tfrc_rx_hist *h, struct sk_buff *skb, u32 n3)
{
u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
s2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_seqno,
s3 = DCCP_SKB_CB(skb)->dccpd_seq;
if (likely(dccp_delta_seqno(s2, s3) > 0)) { /* S2 < S3 */
h->loss_count = 3;
tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 3), skb, n3);
return 1;
}
/* S3 < S2 */
if (dccp_delta_seqno(s1, s3) > 0) { /* S1 < S3 < S2 */
/*
* Reorder history to insert S3 between S1 and S2
*/
tfrc_rx_hist_swap(h, 2, 3);
tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 2), skb, n3);
h->loss_count = 3;
return 1;
}
/* S0 < S3 < S1 */
if (dccp_loss_free(s0, s3, n3)) {
u64 n1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_ndp;
if (dccp_loss_free(s3, s1, n1)) {
/* hole between S0 and S1 filled by S3 */
u64 n2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_ndp;
if (dccp_loss_free(s1, s2, n2)) {
/* entire hole filled by S0, S3, S1, S2 */
tfrc_rx_hist_resume_rtt_sampling(h);
} else {
/* gap remains between S1 and S2 */
h->loss_start = tfrc_rx_hist_index(h, 1);
h->loss_count = 1;
}
} else /* gap exists between S3 and S1, loss_count stays at 2 */
tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_loss_prev(h), skb, n3);
return 0;
}
/*
* The remaining case: S0 < S3 < S1 < S2; gap between S0 and S3
* Reorder history to insert S3 between S0 and S1.
*/
tfrc_rx_hist_swap(h, 0, 3);
h->loss_start = tfrc_rx_hist_index(h, 3);
tfrc_rx_hist_entry_from_skb(tfrc_rx_hist_entry(h, 1), skb, n3);
h->loss_count = 3;
return 1;
}
/* recycle RX history records to continue loss detection if necessary */
static void __three_after_loss(struct tfrc_rx_hist *h)
{
/*
* At this stage we know already that there is a gap between S0 and S1
* (since S0 was the highest sequence number received before detecting
* the loss). To recycle the loss record, it is thus only necessary to
* check for other possible gaps between S1/S2 and between S2/S3.
*/
u64 s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
s2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_seqno,
s3 = tfrc_rx_hist_entry(h, 3)->tfrchrx_seqno;
u64 n2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_ndp,
n3 = tfrc_rx_hist_entry(h, 3)->tfrchrx_ndp;
if (dccp_loss_free(s1, s2, n2)) {
if (dccp_loss_free(s2, s3, n3)) {
/* no gap between S2 and S3: entire hole is filled */
tfrc_rx_hist_resume_rtt_sampling(h);
} else {
/* gap between S2 and S3 */
h->loss_start = tfrc_rx_hist_index(h, 2);
h->loss_count = 1;
}
} else { /* gap between S1 and S2 */
h->loss_start = tfrc_rx_hist_index(h, 1);
h->loss_count = 2;
}
}
/**
* tfrc_rx_congestion_event - Loss detection and further processing
* @h: The non-empty RX history object
* @lh: Loss Intervals database to update
* @skb: Currently received packet
* @ndp: The NDP count belonging to @skb
* @first_li: Caller-dependent computation of first loss interval in @lh
* @sk: Used by @calc_first_li (see tfrc_lh_interval_add)
* Chooses action according to pending loss, updates LI database when a new
* loss was detected, and does required post-processing. Returns 1 when caller
* should send feedback, 0 otherwise.
* Since it also takes care of reordering during loss detection and updates the
* records accordingly, the caller should not perform any more RX history
* operations when loss_count is greater than 0 after calling this function.
*/
bool tfrc_rx_congestion_event(struct tfrc_rx_hist *h,
struct tfrc_loss_hist *lh,
struct sk_buff *skb, const u64 ndp,
u32 (*first_li)(struct sock *), struct sock *sk)
{
bool new_event = false;
if (tfrc_rx_hist_duplicate(h, skb))
return 0;
if (h->loss_count == 0) {
__do_track_loss(h, skb, ndp);
tfrc_rx_hist_sample_rtt(h, skb);
tfrc_rx_hist_add_packet(h, skb, ndp);
} else if (h->loss_count == 1) {
__one_after_loss(h, skb, ndp);
} else if (h->loss_count != 2) {
DCCP_BUG("invalid loss_count %d", h->loss_count);
} else if (__two_after_loss(h, skb, ndp)) {
/*
* Update Loss Interval database and recycle RX records
*/
new_event = tfrc_lh_interval_add(lh, h, first_li, sk);
__three_after_loss(h);
}
/*
* Update moving-average of `s' and the sum of received payload bytes.
*/
if (dccp_data_packet(skb)) {
const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
h->packet_size = tfrc_ewma(h->packet_size, payload, 9);
h->bytes_recvd += payload;
}
/* RFC 3448, 6.1: update I_0, whose growth implies p <= p_prev */
if (!new_event)
tfrc_lh_update_i_mean(lh, skb);
return new_event;
}
EXPORT_SYMBOL_GPL(tfrc_rx_congestion_event);
/* Compute the sending rate X_recv measured between feedback intervals */
u32 tfrc_rx_hist_x_recv(struct tfrc_rx_hist *h, const u32 last_x_recv)
{
u64 bytes = h->bytes_recvd, last_rtt = h->rtt_estimate;
s64 delta = ktime_to_us(net_timedelta(h->bytes_start));
WARN_ON(delta <= 0);
/*
* Ensure that the sampling interval for X_recv is at least one RTT,
* by extending the sampling interval backwards in time, over the last
* R_(m-1) seconds, as per rfc3448bis-06, 6.2.
* To reduce noise (e.g. when the RTT changes often), this is only
* done when delta is smaller than RTT/2.
*/
if (last_x_recv > 0 && delta < last_rtt/2) {
tfrc_pr_debug("delta < RTT ==> %ld us < %u us\n",
(long)delta, (unsigned)last_rtt);
delta = (bytes ? delta : 0) + last_rtt;
bytes += div_u64((u64)last_x_recv * last_rtt, USEC_PER_SEC);
}
if (unlikely(bytes == 0)) {
DCCP_WARN("X_recv == 0, using old value of %u\n", last_x_recv);
return last_x_recv;
}
return scaled_div32(bytes, delta);
}
EXPORT_SYMBOL_GPL(tfrc_rx_hist_x_recv);
void tfrc_rx_hist_purge(struct tfrc_rx_hist *h)
{
int i;
for (i = 0; i <= TFRC_NDUPACK; ++i)
if (h->ring[i] != NULL) {
kmem_cache_free(tfrc_rx_hist_slab, h->ring[i]);
h->ring[i] = NULL;
}
}
EXPORT_SYMBOL_GPL(tfrc_rx_hist_purge);
static int tfrc_rx_hist_alloc(struct tfrc_rx_hist *h)
{
int i;
memset(h, 0, sizeof(*h));
for (i = 0; i <= TFRC_NDUPACK; i++) {
h->ring[i] = kmem_cache_alloc(tfrc_rx_hist_slab, GFP_ATOMIC);
if (h->ring[i] == NULL) {
tfrc_rx_hist_purge(h);
return -ENOBUFS;
}
}
return 0;
}
int tfrc_rx_hist_init(struct tfrc_rx_hist *h, struct sock *sk)
{
if (tfrc_rx_hist_alloc(h))
return -ENOBUFS;
/*
* Initialise first entry with GSR to start loss detection as early as
* possible. Code using this must not use any other fields. The entry
* will be overwritten once the CCID updates its received packets.
*/
tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno = dccp_sk(sk)->dccps_gsr;
return 0;
}
EXPORT_SYMBOL_GPL(tfrc_rx_hist_init);
/**
* tfrc_rx_hist_sample_rtt - Sample RTT from timestamp / CCVal
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. The algorithm is based the concepts presented in RFC 4340, 8.1, using timestamps and the CCVal window counter. There exist 4 cases for the CCVal difference: * == 0: less than RTT/4 passed since last packet -- unusable; * > 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 tried to optimise by storing away the candidate and then re-trying next time. The problem is that * a large number of samples is needed to smooth out the inaccuracies of the algorithm; * the sender may not be sending enough packets to warrant a "next time"; * hence it is better to use suboptimal samples whenever possible. The algorithm now stores away the current sample only if the difference is 0. Applicability and background ---------------------------- A realistic example is MP3 streaming where packets are sent at a rate of less than one packet per RTT, which means that suitable samples are 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%), and 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. Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
2008-09-04 07:30:19 +02:00
* 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)
{
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. The algorithm is based the concepts presented in RFC 4340, 8.1, using timestamps and the CCVal window counter. There exist 4 cases for the CCVal difference: * == 0: less than RTT/4 passed since last packet -- unusable; * > 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 tried to optimise by storing away the candidate and then re-trying next time. The problem is that * a large number of samples is needed to smooth out the inaccuracies of the algorithm; * the sender may not be sending enough packets to warrant a "next time"; * hence it is better to use suboptimal samples whenever possible. The algorithm now stores away the current sample only if the difference is 0. Applicability and background ---------------------------- A realistic example is MP3 streaming where packets are sent at a rate of less than one packet per RTT, which means that suitable samples are 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%), and 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. Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
2008-09-04 07:30:19 +02:00
struct tfrc_rx_hist_entry *last = h->ring[0];
u32 sample, delta_v;
/*
* When not to sample:
* - on non-data packets
* (RFC 4342, 8.1: CCVal only fully defined for data packets);
* - when no data packets have been received yet
* (FIXME: using sampled packet size as indicator here);
* - as long as there are gaps in the sequence space (pending loss).
*/
if (!dccp_data_packet(skb) || h->packet_size == 0 ||
tfrc_rx_hist_loss_pending(h))
return;
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. The algorithm is based the concepts presented in RFC 4340, 8.1, using timestamps and the CCVal window counter. There exist 4 cases for the CCVal difference: * == 0: less than RTT/4 passed since last packet -- unusable; * > 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 tried to optimise by storing away the candidate and then re-trying next time. The problem is that * a large number of samples is needed to smooth out the inaccuracies of the algorithm; * the sender may not be sending enough packets to warrant a "next time"; * hence it is better to use suboptimal samples whenever possible. The algorithm now stores away the current sample only if the difference is 0. Applicability and background ---------------------------- A realistic example is MP3 streaming where packets are sent at a rate of less than one packet per RTT, which means that suitable samples are 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%), and 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. Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
2008-09-04 07:30:19 +02:00
h->rtt_sample_prev = 0; /* reset previous candidate */
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. The algorithm is based the concepts presented in RFC 4340, 8.1, using timestamps and the CCVal window counter. There exist 4 cases for the CCVal difference: * == 0: less than RTT/4 passed since last packet -- unusable; * > 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 tried to optimise by storing away the candidate and then re-trying next time. The problem is that * a large number of samples is needed to smooth out the inaccuracies of the algorithm; * the sender may not be sending enough packets to warrant a "next time"; * hence it is better to use suboptimal samples whenever possible. The algorithm now stores away the current sample only if the difference is 0. Applicability and background ---------------------------- A realistic example is MP3 streaming where packets are sent at a rate of less than one packet per RTT, which means that suitable samples are 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%), and 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. Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
2008-09-04 07:30:19 +02:00
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;
}
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. The algorithm is based the concepts presented in RFC 4340, 8.1, using timestamps and the CCVal window counter. There exist 4 cases for the CCVal difference: * == 0: less than RTT/4 passed since last packet -- unusable; * > 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 tried to optimise by storing away the candidate and then re-trying next time. The problem is that * a large number of samples is needed to smooth out the inaccuracies of the algorithm; * the sender may not be sending enough packets to warrant a "next time"; * hence it is better to use suboptimal samples whenever possible. The algorithm now stores away the current sample only if the difference is 0. Applicability and background ---------------------------- A realistic example is MP3 streaming where packets are sent at a rate of less than one packet per RTT, which means that suitable samples are 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%), and 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. Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
2008-09-04 07:30:19 +02:00
sample = dccp_sane_rtt(ktime_to_us(net_timedelta(last->tfrchrx_tstamp)));
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. The algorithm is based the concepts presented in RFC 4340, 8.1, using timestamps and the CCVal window counter. There exist 4 cases for the CCVal difference: * == 0: less than RTT/4 passed since last packet -- unusable; * > 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 tried to optimise by storing away the candidate and then re-trying next time. The problem is that * a large number of samples is needed to smooth out the inaccuracies of the algorithm; * the sender may not be sending enough packets to warrant a "next time"; * hence it is better to use suboptimal samples whenever possible. The algorithm now stores away the current sample only if the difference is 0. Applicability and background ---------------------------- A realistic example is MP3 streaming where packets are sent at a rate of less than one packet per RTT, which means that suitable samples are 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%), and 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. Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
2008-09-04 07:30:19 +02:00
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;
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. The algorithm is based the concepts presented in RFC 4340, 8.1, using timestamps and the CCVal window counter. There exist 4 cases for the CCVal difference: * == 0: less than RTT/4 passed since last packet -- unusable; * > 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 tried to optimise by storing away the candidate and then re-trying next time. The problem is that * a large number of samples is needed to smooth out the inaccuracies of the algorithm; * the sender may not be sending enough packets to warrant a "next time"; * hence it is better to use suboptimal samples whenever possible. The algorithm now stores away the current sample only if the difference is 0. Applicability and background ---------------------------- A realistic example is MP3 streaming where packets are sent at a rate of less than one packet per RTT, which means that suitable samples are 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%), and 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. Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
2008-09-04 07:30:19 +02:00
/* 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);