TCP拥塞控制算法 — CUBIC的补丁（四）

描述

以下是提交者Stephen Hemminger对这个patch的描述：

enable high resolution ack time if needed

This is a refined version of an earlier patch by Lucas Nussbaum.

Cubic needs RTT values in miliseconds. If HZ < 1000 then the values will be too coarse.

代码

--- a/net/ipv4/tcp_cubic.c
+++ b/net/ipv4/tcp_cubic.c
@@ -459,6 +459,10 @@ static int __init cubictcp_register(void)
/* divide by bic_scale and by constant Srtt (100ms) */
do_div(cube_factor, bic_scale * 10);

+	/* hystart needs ms clock resolution */
+	if (hystart && HZ < 1000)
+		cubictcp.flags |= TCP_CONG_RTT_STAMP;
+
return tcp_register_congestion_control(&cubictcp);
}

分析

@include/net/tcp.h：
#define TCP_CONG_RTT_STAMP 0x02

struct tcp_congestion_ops {
...
unsigned long flags;
...
};

/**
* @tstamp: Time we arrived
*/
struct sk_buff {
...
ktime_t tstamp;
...
};

static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets, u32 prior_snd_una)
{
ktime_t last_ackt = net_invalid_timestamp();
...
ca_seq_rtt = now - scb->when;
last_ackt = skb->tstamp;
...
if (ca_ops->pkts_acked) {
s32 rtt_us = -1;

/* Is the ACK triggering packet unambiguous? */
if (! (flag & FLAG_RETRANS_DATA_ACKED)) {
/* High resolution needed and available? */
if (ca_ops->flags & TCP_CONG_RTT_STAMP &&
! ktime_equal(last_ackt, net_invalid_timestamp()))
rtt_us = ktime_us_delta(ktime_get_real(), last_ackt); /* 精确到微秒*/

else if (ca_seq_rtt >= 0)
rtt_us = jiffies_to_usecs(ca_seq_rtt); /*精确度只有jiffies，一般是毫秒 */
}
ca_ops->pkts_acked(sk, pkts_acked, rtt_us);
}
...
}

/**
* ktime_equal - Compares two ktime_t variables to see if they are equal
* @cmp1: comparable1
* @cmp2: comparable2
* Compare two ktime_t variables, returns 1 if equal
*/
static inline int ktime_equal(const ktime_t cmp1, const ktime_t cmp2)
{
return cmp1.tv64 == cmp2.tv64;
}

static inline ktime_t net_invalid_timestamp(void)
{
return ktime_set(0, 0);
}

/* Set a ktime_t variable to a value in sec/nsec representation: */
static inline ktime_t ktime_set (const long secs, const unsigned long nsecs)
{
return (ktime_t) { .tv = { .sec = secs, .nsec = nsecs} };
}

@tcp_transmit_skb
/* If congestion control is doing timestamping, we must take such a timestamp
* before we potentially clone/copy.
*/
if (icsk->icsk_ca_ops->flags & TCP_CONG_RTT_STAMP)
__net_timestamp(skb); /*设置skb->tstamp，记录发送时间*/

static inline void __net_timestamp(struct sk_buff *skb)
{
skb->tstamp = ktime_get_real(); /* 获取当前时间的ktime_t表示 */
}

评价

我们知道当HZ < 1000时，计算得到的RTT的精确度就在1ms以上。这个时候可以通过设置

TCP_CONG_RTT_STAMP标志来获得更高精度的RTT，测量得到的RTT可以精确到微秒。

但是由于x86机器的HZ一般为1000，所以这个补丁的影响有限。

Author

zhangskd @ csdn blog