TCP being hoodwinked into spurious retransmissions by lack of timestamps?
Subsystems (3)
I've been looking at some packet traces of an application looking to upload a Large Quantity (tm) of data to a server across the Big Bad Internet (tm). They've been Linux senders, and the destination while supporting SACK and window scaling does not support TCP timestamps. (TCP timestamp support was requested of the supplier of said server many many months ago now.)
This destination system has been issuing RSTs at seemingly random points in the middle of a large fraction of the attempted transfers. In looking at the traces, they all seem to be variations on the theme of what is shown by:
ftp://netperf.org/retrans_question/for_netdev.png
which is a passing of ftp://netperf.org/retrans_question/for_netdev.pcap through tcptrace -nG and zoomed-in to the end. I've seen this with a 3.2.0 kernel as the sender, have reports of it happening with whatever is in Fedora Core 20, and the traces above are from a 3.11.0 kernel as the sender.
The large quantity of (likely) unnecessary retransmissions shouldn't be triggering a RST by the receiver, but the failures consistently show that and I was wondering if the (spurious) retransmissions were perhaps "encouraged" (so to speak) by the lack of TCP Timestamps.
happy benchmarking,
rick jones
Running with such a large window scale and no timestamps (PAWS protection) is generally not a great idea, but I don't think is part of the issue here.
If you look where things really go wrong, the receiver is sending anomalous SACK blocks that will trigger the SACK renege handling path.
Reneging triggers go-back-n behavior, so we see the spurious
retransmits from there on.
The most notable bad segment is this one:
18:20:46.800063 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3171368, win 32716, options [nop,nop,sack 1 {3171368:3177208}], length 0
It contains a SACK block contiguous with the acked seqno. There is some other strangeness just before that, where the SACK block shrinks then grows again.
One other thing that jumped out at me is there is no actual loss, just reordering.
-John
Quoted 29 lines · Rick Jones
I've been looking at some packet traces of an application looking to upload a Large Quantity (tm) of data to a server across the Big Bad Internet (tm). They've been Linux senders, and the destination while supporting SACK and window scaling does not support TCP timestamps. (TCP timestamp support was requested of the supplier of said server many many months ago now.)
This destination system has been issuing RSTs at seemingly random points in the middle of a large fraction of the attempted transfers. In looking at the traces, they all seem to be variations on the theme of what is shown by:
ftp://netperf.org/retrans_question/for_netdev.png
which is a passing of ftp://netperf.org/retrans_question/for_netdev.pcap through tcptrace -nG and zoomed-in to the end. I've seen this with a 3.2.0 kernel as the sender, have reports of it happening with whatever is in Fedora Core 20, and the traces above are from a 3.11.0 kernel as the sender.
The large quantity of (likely) unnecessary retransmissions shouldn't be triggering a RST by the receiver, but the failures consistently show that and I was wondering if the (spurious) retransmissions were perhaps "encouraged" (so to speak) by the lack of TCP Timestamps.
happy benchmarking,
rick jones
--
To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Quoted 3 lines · John Heffner
Running with such a large window scale and no timestamps (PAWS protection) is generally not a great idea, but I don't think is part of the issue here.
OK. I'll look for other, additional sticks with which to beat the provider of the system that doesn't do timestamping :)
Quoted 4 lines
If you look where things really go wrong, the receiver is sending anomalous SACK blocks that will trigger the SACK renege handling path.
Reneging triggers go-back-n behavior, so we see the spurious
retransmits from there on.
Should those subsequent ACKs be clocking-out additional retransmissions like they appear to do? (Assuming I'm not projecting into the trace) Or is that an unavoidable consequence of there being no timestamps with which to tell which send was being ACKed?
Quoted 5 lines
The most notable bad segment is this one:
18:20:46.800063 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3171368, win 32716, options [nop,nop,sack 1 {3171368:3177208}], length 0
It contains a SACK block contiguous with the acked seqno.
I've gone back through one of the other traces and found the same thing therein.
Quoted 2 lines
There is some other strangeness just before that, where the SACK block shrinks then grows again.
That would be this yes?
15:20:46.798816 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3660468:3661928, ack 4262, win 297, length 1460 15:20:46.799027 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3168256, win 32081, options [nop,nop,sack 1 {3171368:3172828}], length 0 15:20:46.799042 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3661928:3664848, ack 4262, win 297, length 2920 15:20:46.799465 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3169716, win 32241, options [nop,nop,sack 1 {3171368:3172828}], length 0 15:20:46.799479 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3664848:3666308, ack 4262, win 297, length 1460 15:20:46.799497 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3169716, win 32241, options [nop,nop,sack 1 {3171368:3174288}], length 0 15:20:46.799504 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3169716, win 32241, options [nop,nop,sack 1 {3171368:3175748}], length 0 15:20:46.799509 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3666308:3667768, ack 4262, win 297, length 1460 15:20:46.799773 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3171176, win 32491, options [nop,nop,sack 1 {3171368:3172828}], length 0 15:20:46.799787 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3667768:3669228, ack 4262, win 297, length 1460 15:20:46.800063 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3171368, win 32716, options [nop,nop,sack 1 {3171368:3177208}], length 0 15:20:46.800081 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3171368:3172828, ack 4262, win 297, length 1460
Might that be packet-reordering in the other direction? Sadly, I don't have good "both sides" traces as the receiving system doesn't seem to capture traffic terribly well. I suppose TCP timestamps might have helped answer that question.
Quoted 2 lines
One other thing that jumped out at me is there is no actual loss, just reordering.
That was interesting wasn't it - calling it the "Big Bad Internet (tm)" I was expecting to see *some* actual packet loss. Though in this case the traffic, while going across the continental US, may not have been crossing Internet providers.
thanks muchly,
rick jones
Quoted 38 lines
-John
Quoted 22 lines · Rick Jones
I've been looking at some packet traces of an application looking to upload a Large Quantity (tm) of data to a server across the Big Bad Internet (tm). They've been Linux senders, and the destination while supporting SACK and window scaling does not support TCP timestamps. (TCP timestamp support was requested of the supplier of said server many many months ago now.)
This destination system has been issuing RSTs at seemingly random points in the middle of a large fraction of the attempted transfers. In looking at the traces, they all seem to be variations on the theme of what is shown by:
ftp://netperf.org/retrans_question/for_netdev.png
which is a passing of ftp://netperf.org/retrans_question/for_netdev.pcap through tcptrace -nG and zoomed-in to the end. I've seen this with a 3.2.0 kernel as the sender, have reports of it happening with whatever is in Fedora Core 20, and the traces above are from a 3.11.0 kernel as the sender.
The large quantity of (likely) unnecessary retransmissions shouldn't be triggering a RST by the receiver, but the failures consistently show that and I was wondering if the (spurious) retransmissions were perhaps "encouraged" (so to speak) by the lack of TCP Timestamps.
Quoted 7 lines
happy benchmarking,
rick jones
--
To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html--
To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Quoted 11 lines · Rick Jones
Quoted 4 lines · John Heffner
If you look where things really go wrong, the receiver is sending anomalous SACK blocks that will trigger the SACK renege handling path.
Reneging triggers go-back-n behavior, so we see the spurious
retransmits from there on.Should those subsequent ACKs be clocking-out additional retransmissions like they appear to do? (Assuming I'm not projecting into the trace) Or is that an unavoidable consequence of there being no timestamps with which to tell which send was being ACKed?
It's possible that if they had timestamps, the ACKs for for original transmits might not open up the window, thus preventing the retransmits from going out. But I'm not nearly familiar enough with the details of the current retransmit machinery to know for sure. Reneging like this (go-back-n before a timeout) is not a commonly-exercised case, so I wouldn't be too surprised if the sender isn't doing the smartest possible thing.
Quoted 42 lines
Quoted 5 lines
The most notable bad segment is this one:
18:20:46.800063 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3171368, win 32716, options [nop,nop,sack 1 {3171368:3177208}], length 0
It contains a SACK block contiguous with the acked seqno.I've gone back through one of the other traces and found the same thing therein.
Quoted 2 lines
There is some other strangeness just before that, where the SACK block shrinks then grows again.
That would be this yes?
15:20:46.798816 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3660468:3661928, ack 4262, win 297, length 1460 15:20:46.799027 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3168256, win 32081, options [nop,nop,sack 1 {3171368:3172828}], length 0 15:20:46.799042 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3661928:3664848, ack 4262, win 297, length 2920 15:20:46.799465 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3169716, win 32241, options [nop,nop,sack 1 {3171368:3172828}], length 0 15:20:46.799479 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3664848:3666308, ack 4262, win 297, length 1460 15:20:46.799497 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3169716, win 32241, options [nop,nop,sack 1 {3171368:3174288}], length 0 15:20:46.799504 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3169716, win 32241, options [nop,nop,sack 1 {3171368:3175748}], length 0 15:20:46.799509 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3666308:3667768, ack 4262, win 297, length 1460 15:20:46.799773 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3171176, win 32491, options [nop,nop,sack 1 {3171368:3172828}], length 0 15:20:46.799787 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3667768:3669228, ack 4262, win 297, length 1460 15:20:46.800063 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3171368, win 32716, options [nop,nop,sack 1 {3171368:3177208}], length 0 15:20:46.800081 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3171368:3172828, ack 4262, win 297, length 1460
Quoted 4 lines
Might that be packet-reordering in the other direction? Sadly, I don't have good "both sides" traces as the receiving system doesn't seem to capture traffic terribly well. I suppose TCP timestamps might have helped answer that question.
Regardless of any possible reordering, in this case we know something odd is going on in the receiver because ACK advances at the same time the SACK block shrinks.
-John
Quoted 31 lines
Quoted 31 lines
Quoted 2 lines
There is some other strangeness just before that, where the SACK block shrinks then grows again.
That would be this yes?
15:20:46.798816 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3660468:3661928, ack 4262, win 297, length 1460 15:20:46.799027 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3168256, win 32081, options [nop,nop,sack 1 {3171368:3172828}], length 0 15:20:46.799042 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3661928:3664848, ack 4262, win 297, length 2920 15:20:46.799465 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3169716, win 32241, options [nop,nop,sack 1 {3171368:3172828}], length 0 15:20:46.799479 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3664848:3666308, ack 4262, win 297, length 1460 15:20:46.799497 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3169716, win 32241, options [nop,nop,sack 1 {3171368:3174288}], length 0 15:20:46.799504 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3169716, win 32241, options [nop,nop,sack 1 {3171368:3175748}], length 0 15:20:46.799509 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3666308:3667768, ack 4262, win 297, length 1460 15:20:46.799773 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3171176, win 32491, options [nop,nop,sack 1 {3171368:3172828}], length 0 15:20:46.799787 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3667768:3669228, ack 4262, win 297, length 1460 15:20:46.800063 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3171368, win 32716, options [nop,nop,sack 1 {3171368:3177208}], length 0 15:20:46.800081 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3171368:3172828, ack 4262, win 297, length 1460
Quoted 8 lines
Quoted 4 lines
Might that be packet-reordering in the other direction? Sadly, I don't have good "both sides" traces as the receiving system doesn't seem to capture traffic terribly well. I suppose TCP timestamps might have helped answer that question.
Regardless of any possible reordering, in this case we know something odd is going on in the receiver because ACK advances at the same time the SACK block shrinks.
Ah yes, I'd not picked-up on that.
thanks,
rick jones
Neal Cardwell
maintainerWhat's the receiver OS in this trace? It's reneging on SACKs. :-) Take a look at this ACK:
18:20:46.800063 IP 75.236.145.7.443 > 91.216.86.7.56064: . 4262:4262(0) ack 3171368 win 32716 <nop,nop,sack 1 {3171368:3177208}>
Note that it's ACKing 3171368 and SACKing the adjacent sequence range:
{3171368:3177208}. That's not cool.
I think that's causing the Linux sender to enter the tcp_check_sack_reneging() code path, which calls tcp_enter_loss().
It seems that the Linux sender did not enable FRTO for that tcp_enter_loss() invocation. Maybe there is some way we can revise the logic to enable FRTO in cases like this, so we can detect that the retransmission was not needed, and abort the stream of spurious retransmissions...
neal
Quoted 55 lines · Rick Jones
Quoted 39 lines
Quoted 32 lines
Quoted 2 lines
There is some other strangeness just before that, where the SACK block shrinks then grows again.
That would be this yes?
15:20:46.798816 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3660468:3661928, ack 4262, win 297, length 1460 15:20:46.799027 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3168256, win 32081, options [nop,nop,sack 1 {3171368:3172828}], length 0 15:20:46.799042 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3661928:3664848, ack 4262, win 297, length 2920 15:20:46.799465 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3169716, win 32241, options [nop,nop,sack 1 {3171368:3172828}], length 0 15:20:46.799479 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3664848:3666308, ack 4262, win 297, length 1460 15:20:46.799497 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3169716, win 32241, options [nop,nop,sack 1 {3171368:3174288}], length 0 15:20:46.799504 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3169716, win 32241, options [nop,nop,sack 1 {3171368:3175748}], length 0 15:20:46.799509 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3666308:3667768, ack 4262, win 297, length 1460 15:20:46.799773 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3171176, win 32491, options [nop,nop,sack 1 {3171368:3172828}], length 0 15:20:46.799787 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3667768:3669228, ack 4262, win 297, length 1460 15:20:46.800063 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3171368, win 32716, options [nop,nop,sack 1 {3171368:3177208}], length 0 15:20:46.800081 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3171368:3172828, ack 4262, win 297, length 1460
Quoted 5 lines
Might that be packet-reordering in the other direction? Sadly, I don't have
good "both sides" traces as the receiving system doesn't seem to capture traffic terribly well. I suppose TCP timestamps might have helped answer that question.Quoted 3 lines
Regardless of any possible reordering, in this case we know something odd is going on in the receiver because ACK advances at the same time the SACK block shrinks.
Ah yes, I'd not picked-up on that.
thanks,
rick jones
--
To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Quoted 1 line · Neal Cardwell
What's the receiver OS in this trace?
It is my understanding that the receiver has a FreeBSD-based stack.
rick
Quoted 9 lines · Neal Cardwell
What's the receiver OS in this trace? It's reneging on SACKs. :-) Take a look at this ACK:
18:20:46.800063 IP 75.236.145.7.443 > 91.216.86.7.56064: . 4262:4262(0) ack 3171368 win 32716 <nop,nop,sack 1 {3171368:3177208}>
Note that it's ACKing 3171368 and SACKing the adjacent sequence range:
{3171368:3177208}. That's not cool.
Quoted 8 lines
I think that's causing the Linux sender to enter the
tcp_check_sack_reneging()code path, which callstcp_enter_loss().It seems that the Linux sender did not enable FRTO for that
tcp_enter_loss()invocation. Maybe there is some way we can revise the logic to enable FRTO in cases like this, so we can detect that the retransmission was not needed, and abort the stream of spurious retransmissions...
Sure we can try:
net/ipv4/tcp_input.c +1 -1
@@ -1972,7 +1972,7 @@ void tcp_enter_loss(struct sock *sk, int how)* the same SND.UNA (sec 3.2). Disable F-RTO on path MTU probing */ tp->frto = sysctl_tcp_frto && - (new_recovery || icsk->icsk_retransmits) && + (new_recovery || icsk->icsk_retransmits || how) && !inet_csk (sk)->icsk_mtup.probe_size; }
However that only works if we got new data to send. For a better solution, with the lack of TS option or DSACK support, we can
1) use Neal's neat idea to send a different size packet on the first retransmission after timeout, and use that to distinguish if the ACK is for the original or retry.
2) Do not blindly marked any packet unsacked lost in tcp_enter_loss; My idea would be to do that only if the packet was sent min_rtt ago;
I can try to implement these ideas if people are interested.
Quoted 60 lines
neal
Quoted 55 lines · Rick Jones
Quoted 39 lines
Quoted 32 lines
Quoted 2 lines
There is some other strangeness just before that, where the SACK block shrinks then grows again.
That would be this yes?
15:20:46.798816 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3660468:3661928, ack 4262, win 297, length 1460 15:20:46.799027 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3168256, win 32081, options [nop,nop,sack 1 {3171368:3172828}], length 0 15:20:46.799042 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3661928:3664848, ack 4262, win 297, length 2920 15:20:46.799465 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3169716, win 32241, options [nop,nop,sack 1 {3171368:3172828}], length 0 15:20:46.799479 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3664848:3666308, ack 4262, win 297, length 1460 15:20:46.799497 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3169716, win 32241, options [nop,nop,sack 1 {3171368:3174288}], length 0 15:20:46.799504 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3169716, win 32241, options [nop,nop,sack 1 {3171368:3175748}], length 0 15:20:46.799509 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3666308:3667768, ack 4262, win 297, length 1460 15:20:46.799773 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3171176, win 32491, options [nop,nop,sack 1 {3171368:3172828}], length 0 15:20:46.799787 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3667768:3669228, ack 4262, win 297, length 1460 15:20:46.800063 IP 75.236.145.7.443 > 91.216.86.7.56064: Flags [.], ack 3171368, win 32716, options [nop,nop,sack 1 {3171368:3177208}], length 0 15:20:46.800081 IP 91.216.86.7.56064 > 75.236.145.7.443: Flags [.], seq 3171368:3172828, ack 4262, win 297, length 1460
Quoted 5 lines
Might that be packet-reordering in the other direction? Sadly, I don't have
good "both sides" traces as the receiving system doesn't seem to capture traffic terribly well. I suppose TCP timestamps might have helped answer that question.Quoted 3 lines
Regardless of any possible reordering, in this case we know something odd is going on in the receiver because ACK advances at the same time the SACK block shrinks.
Ah yes, I'd not picked-up on that.
thanks,
rick jones
--
To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Quoted 34 lines · Yuchung Cheng
Quoted 9 lines · Neal Cardwell
What's the receiver OS in this trace? It's reneging on SACKs. :-) Take a look at this ACK:
18:20:46.800063 IP 75.236.145.7.443 > 91.216.86.7.56064: . 4262:4262(0) ack 3171368 win 32716 <nop,nop,sack 1 {3171368:3177208}>
Note that it's ACKing 3171368 and SACKing the adjacent sequence range:
{3171368:3177208}. That's not cool.Quoted 8 lines
I think that's causing the Linux sender to enter the
tcp_check_sack_reneging()code path, which callstcp_enter_loss().It seems that the Linux sender did not enable FRTO for that
tcp_enter_loss()invocation. Maybe there is some way we can revise the logic to enable FRTO in cases like this, so we can detect that the retransmission was not needed, and abort the stream of spurious retransmissions...Sure we can try:
net/ipv4/tcp_input.c +1
-1@@ -1972,7 +1972,7 @@ void tcp_enter_loss(struct sock *sk, int how)* the same SND.UNA (sec 3.2). Disable F-RTO on path MTU probing */ tp->frto = sysctl_tcp_frto && - (new_recovery || icsk->icsk_retransmits) && + (new_recovery || icsk->icsk_retransmits || how) && !inet_csk (sk)->icsk_mtup.probe_size; }
Quoted 5 lines
However that only works if we got new data to send. For a better solution, with the lack of TS option or DSACK support, we can
1) use Neal's neat idea to send a different size packet on the first retransmission after timeout, and use that to distinguish if the ACK is for the original or retry.
What would one do if the ACK arriving after the short retransmission was farther to the right than the end of the original packet? Won't that be ambiguous?
Quoted 4 lines
2) Do not blindly marked any packet unsacked lost in
tcp_enter_loss; My idea would be to do that only if the packet was sentmin_rttago;I can try to implement these ideas if people are interested.
If these near-heroics are unnecessary if timestamps are present, I'm not sure I'd push too hard. Unless you think that timestamps not being present is sufficiently common.
rick
Quoted 8 lines · John Heffner
Running with such a large window scale and no timestamps (PAWS protection) is generally not a great idea, but I don't think is part of the issue here.
If you look where things really go wrong, the receiver is sending anomalous SACK blocks that will trigger the SACK renege handling path.
Reneging triggers go-back-n behavior, so we see the spurious
retransmits from there on.
What triggers go-back-n when SACK is not in use? I ask because at least once I have seen the same sort of thing without SACK enabled on the connection. The total quantity of retransmissions is roughly the same, but spread-out - looks like cwnd shrinks considerably and re-grows in the no-SACK case. Not sure I still have that trace though :(
rick jones
Quoted 20 lines · Rick Jones
I've been looking at some packet traces of an application looking to upload a Large Quantity (tm) of data to a server across the Big Bad Internet (tm). They've been Linux senders, and the destination while supporting SACK and window scaling does not support TCP timestamps. (TCP timestamp support was requested of the supplier of said server many many months ago now.)
This destination system has been issuing RSTs at seemingly random points in the middle of a large fraction of the attempted transfers. In looking at the traces, they all seem to be variations on the theme of what is shown by:
ftp://netperf.org/retrans_question/for_netdev.png
which is a passing of ftp://netperf.org/retrans_question/for_netdev.pcap through tcptrace -nG and zoomed-in to the end. I've seen this with a 3.2.0 kernel as the sender, have reports of it happening with whatever is in Fedora Core 20, and the traces above are from a 3.11.0 kernel as the sender.
Quoted 4 lines
The large quantity of (likely) unnecessary retransmissions shouldn't be triggering a RST by the receiver, but the failures consistently show that and I was wondering if the (spurious) retransmissions were perhaps "encouraged" (so to speak) by the lack of TCP Timestamps.
I have learned why the receiving TCP has reset the connection. It would seem that stack has a heuristic whereby if it receives more than 255 retransmissions in a window it will abort the connection.
rick jones