a F-RTO question

DISCUSSION Li Yu 4 participants 7 messages Mar 28, 2012 – Apr 10, 2012

Subsystems (2)

Networking Network drivers

lore

Share & tools

lore

Li Yu

author

2012-03-28 03:14 UTC #0 lore raw

Hi,

I have a question about tcp_process_frto(), the below source
code :

static int tcp_process_frto(struct sock *sk, int flag)

{
	.....

        if (!before(tp->snd_una, tp->frto_highmark)) {
                tcp_enter_frto_loss(sk, ...);
                return 1;
        }

	.....

}

	As my understanding, the tp->frto_highmark likes tp->high_seq,

it saves the seqno SND_NXT when a TCP connection enters F-RTO phase, is it the variable "recovery" in NewReno? So I think that if snd_una is equal with or after frto_highmark, which means peer ack new data, so why we enter Loss state here?

Thanks!

Chao Pei

2012-03-28 03:49 UTC #1 lore raw

Quoted 24 lines

Hi,
I have a question about tcp_process_frto(), the below source code :
static int tcp_process_frto(struct sock *sk, int flag) {
.....
if (!before(tp->snd_una, tp->frto_highmark)) { tcp_enter_frto_loss(sk, ...); return 1;
}
.....
}
As my understanding, the tp->frto_highmark likes tp->high_seq, it saves the seqno SND_NXT when a TCP connection enters F-RTO phase, is it the variable "recovery" in NewReno? So I think that if snd_una is equal with or after frto_highmark, which means peer ack new data, so why we enter Loss state here?

Quoted 5 lines

Thanks!
Yu

If snd_una advances to frto_highmark, it is likely that the hole was filled by the retransimitted packet, which means the original packet was likely to have been lost.
So, we should enter loss state.

Li Yu

author

2012-03-28 04:08 UTC #2 lore raw

Quoted 26 lines · Chao Pei

Quoted 26 lines
Hi,
I have a question about tcp_process_frto(), the below source
code :
static int tcp_process_frto(struct sock *sk, int flag)
{
        .....

        if (!before(tp->snd_una, tp->frto_highmark)) {
                tcp_enter_frto_loss(sk, ...);
                return 1;
        }

        .....
}
        As my understanding, the tp->frto_highmark likes tp->high_seq,
it saves the seqno SND_NXT when a TCP connection enters F-RTO phase, is it the variable "recovery" in NewReno? So I think that if snd_una is equal with or after frto_highmark, which means peer ack new data, so why we enter Loss state here?
Thanks!

Quoted 9 lines

Quoted 3 lines
Yu
If snd_una advances to frto_highmark, it is likely that the hole was filled by the retransimitted packet, which means the original packet was likely to have been lost.
So, we should enter loss state.

I do not agree with it, if snd_una advanced to frto_highmark, which means peer acks whole window of data instead of just one segment, and we can not make sure that reason of peer sends ack is whether it received original segment or retransmitted segment.

Even, the reason is latter, it also means the netowrk already is
recovered from temporarily congestion or disordered state, so we also should not enter loss state.

Thanks

Chao Pei

2012-03-28 05:27 UTC #3 lore raw

Quoted 42 lines

Quoted 27 lines
Quoted 27 lines
Hi,
I have a question about tcp_process_frto(), the below source code :
static int tcp_process_frto(struct sock *sk, int flag) {
.....
if (!before(tp->snd_una, tp->frto_highmark)) { tcp_enter_frto_loss(sk, ...); return 1;
}
.....
}
As my understanding, the tp->frto_highmark likes tp->high_seq, it saves the seqno SND_NXT when a TCP connection enters F-RTO phase, is it the variable "recovery" in NewReno? So I think that if snd_una is equal with or after frto_highmark, which means peer ack new data, so why we enter Loss state here?
Thanks!
Quoted 9 lines
Quoted 3 lines
Yu
If snd_una advances to frto_highmark, it is likely that the hole was filled by the retransimitted packet, which means the original packet was likely to have been lost.
So, we should enter loss state.
I do not agree with it, if snd_una advanced to frto_highmark, which means peer acks whole window of data instead of just one segment, and we can not make sure that reason of peer sends ack is whether it received original segment or retransmitted segment.

Quoted 8 lines

Even, the reason is latter, it also means the netowrk already is recovered from temporarily congestion or disordered state, so we also should not enter loss state.
Thanks
Yu

If it is for the first reason, then the receiver should not ack the whole rtt of packets with just one ack. Instead, it should send ack once every second packet. So, we can almost be sure that the original packet was lost.

As for the second question. I think since the loss is proven and was actually detected because of the RTO timer, we should enter loss state.

Thanks.

Yuchung Cheng

2012-03-28 05:35 UTC #4 lore raw

In reply to Li Yu

Quoted 38 lines · Li Yu

Quoted 26 lines · Chao Pei
Quoted 26 lines
Hi,
I have a question about tcp_process_frto(), the below source
code :
static int tcp_process_frto(struct sock *sk, int flag)
{
       .....

       if (!before(tp->snd_una, tp->frto_highmark)) {
               tcp_enter_frto_loss(sk, ...);
               return 1;
       }

       .....
}
       As my understanding, the tp->frto_highmark likes tp->high_seq,
it saves the seqno SND_NXT when a TCP connection enters F-RTO phase, is it the variable "recovery" in NewReno? So I think that if snd_una is equal with or after frto_highmark, which means peer ack new data, so why we enter Loss state here?
Thanks!
Quoted 9 lines
Quoted 3 lines
Yu
If snd_una advances to frto_highmark, it is likely that the hole was filled by the retransimitted packet, which means the original packet was likely to have been lost.
So, we should enter loss state.

Quoted 8 lines

I do not agree with it, if snd_una advanced to frto_highmark, which means peer acks whole window of data instead of just one segment, and we can not make sure that reason of peer sends ack is whether it received original segment or retransmitted segment.
Even, the reason is latter, it also means the netowrk already is recovered from temporarily congestion or disordered state, so we also should not enter loss state.

Like you said there is some ambiguity and F-RTO takes the conservative approach. You can find answers to your question in RFC 5682 (Section 2.2). Unless the RTO is proven to be spurious, TCP should reduce window and performs slow-start regardlessly.

Li Yu

author

2012-03-28 06:43 UTC #5 lore raw

Quoted 44 lines · Yuchung Cheng

Quoted 43 lines · Li Yu
Quoted 26 lines · Chao Pei
Quoted 26 lines
Hi,
I have a question about tcp_process_frto(), the below source
code :
static int tcp_process_frto(struct sock *sk, int flag)
{
        .....

        if (!before(tp->snd_una, tp->frto_highmark)) {
                tcp_enter_frto_loss(sk, ...);
                return 1;
        }

        .....
}
        As my understanding, the tp->frto_highmark likes tp->high_seq,
it saves the seqno SND_NXT when a TCP connection enters F-RTO phase, is it the variable "recovery" in NewReno? So I think that if snd_una is equal with or after frto_highmark, which means peer ack new data, so why we enter Loss state here?
Thanks!
Quoted 9 lines
Quoted 3 lines
Yu
If snd_una advances to frto_highmark, it is likely that the hole was filled by the retransimitted packet, which means the original packet was likely to have been lost.
So, we should enter loss state.
I do not agree with it, if snd_una advanced to frto_highmark, which means peer acks whole window of data instead of just one segment, and we can not make sure that reason of peer sends ack is whether it received original segment or retransmitted segment.

Quoted 8 lines

Quoted 3 lines
Even, the reason is latter, it also means the netowrk already is recovered from temporarily congestion or disordered state, so we also should not enter loss state.
Like you said there is some ambiguity and F-RTO takes the conservative approach. You can find answers to your question in RFC 5682 (Section 2.2). Unless the RTO is proven to be spurious, TCP should reduce window and performs slow-start regardlessly.

I think I got it, if the new ack covers "frto_highmark", it may mean the fast retransmitted segments are lost likely.

In RFC5682:

If the first acknowledgment after the RTO retransmission covers the "recover" point at algorithm step (2a), there is not enough evidence that a non-retransmitted segment has arrived at the receiver after the timeout. This is a common case when a fast retransmission is lost and has been retransmitted again after an RTO, while the rest of the unacknowledged segments were successfully delivered to the TCP receiver before the retransmission timeout. Therefore, the timeout cannot be declared spurious in this case.

And as Chao's words, if the RTO is proven, we should enter slow-start then.

Thanks!

Ilpo Järvinen

2012-04-10 08:39 UTC #6 lore raw

In reply to Li Yu

Quoted 3 lines · Li Yu

Even, the reason is latter, it also means the netowrk already is recovered from temporarily congestion or disordered state, so we also should not enter loss state.

There are also other considerations in this btw, some devices are simply broken and go to RTO loop increasing the RTOs exponentially if we wouldn't force retransmit of the next segment (one of those newly sent segments) that isn't strictly mandatory imho. ...We used to do something more clever here and avoided that retransmission but learned the hard way that there's simply no way around it.

Signature · 1 line

i.