[En-Nut-Discussion] unreliable tcp
Szemző András
saam at kometa.hu
Wed Apr 9 10:47:06 CEST 2008
Hi,
It seems increasing the lower boundary of the tcp retransmission time from
200ms to 2sec solved the problem,
as the modul is online after 17h, without a single reconnect, compared with
the 10-30min, without this
modification in NutTcpCalcRtt() in tcputil.c
Regards,
Andras
-----Original Message-----
From: en-nut-discussion-bounces at egnite.de
[mailto:en-nut-discussion-bounces at egnite.de] On Behalf Of Szemző András
Sent: Tuesday, April 08, 2008 5:29 PM
To: 'Ethernut User Chat (English)'
Subject: Re: [En-Nut-Discussion] unreliable tcp
Hi guys,
I captured the error situation and made a picture about it, maybe it will
help somebody with better tcp/ip knowledge than I have to track down what is
going on here :)
http://www.esh.hu/nut_err.jpg
(blue is the nut board, red is the linux server app)
Thanks a lot
Andras
-----Original Message-----
From: en-nut-discussion-bounces at egnite.de
[mailto:en-nut-discussion-bounces at egnite.de] On Behalf Of Szemző András
Sent: Tuesday, April 08, 2008 2:44 PM
To: 'Ethernut User Chat (English)'
Subject: Re: [En-Nut-Discussion] unreliable tcp
Harald, thanks your answer.
Ok, it seems my english was not enough good to explain my problem completly
:)
Here is a connect handler,log sender and reader part of the main() thread:
// Disconnected state, try to connect to Server
if ( Timer.Num[TIMERCONNECT] == 0 ) {
if ( SkyConnect(NetStatus.UsedServer) ==
SKY_CONNECT_OK ) {
NetStatus.TCPConnectCount=0;
Timer.Num[TIMERCONNECT] = -1;
Timer.Num[TIMERLOGALIVE] =
MainConfig.TcpTestTime;
NetStatus.TcpConnectionState =
TCP_CONNECTED;
if ( NetStatus.UsedServer != 0 ) {
Timer.Num[PRIMARYIPRESTORE] = 900;
} else {
Timer.Num[PRIMARYIPRESTORE] = -1;
}
} else {
NetStatus.TCPConnectCount++;
Timer.Num[TIMERCONNECT] =
CfgTCPConnectWaitTime;
}
}
....
// Log sender
if ( NetStatus.TcpConnectionState == TCP_CONNECTED ) {
rc = SendELogPacket();
if ( rc == -1) {
Timer.Num[TIMERRESENDLOG] =
CfgTCPResendLogTime; // reload timer
NetStatus.TCPResendCount++;
#ifdef _MODUL_DEBUG_
puts("LOG_SEND_ERR");
fflush(stdout);
#endif
}
if ( rc == 1 ) {
#ifdef _MODUL_DEBUG_
puts("LOG_SEND_OK"); fflush(stdout);
#endif
Timer.Num[TIMERRESENDLOG] = -1;
NetStatus.TCPResendCount = 0;
}
if ( StatusSend == ENABLED ) {
StatusSend = DISABLED;
SkySendStatus(TCP_PACKET_STATUS_NEW,
GetRandomNum());
#ifdef _MODUL_DEBUG_
printf("Statussend\r\n");
fflush(stdout);
#endif
}
}
// TCP reader
if ( NetStatus.TcpConnectionState == TCP_CONNECTED ) {
tmo=100;
NutTcpSetSockOpt(sock, SO_RCVTIMEO, &tmo,
sizeof(tmo));
rc = NutTcpReceive(sock, receivebuff,
sizeof(receivebuff)); // see if we get
something
if ( rc == 16) {
NetStatusIn(receivebuff);
} else {
if ( rc == -1) {
// Cannot read socket
#ifdef _MODUL_DEBUG_
printf("READ ERROR,
Disconnected\n"); fflush(stdout);
#endif
SkyDisconnect();
NetStatus.TcpConnectionState =
TCP_DISCONNECTED;
Timer.Num[TIMERCONNECT] =
CfgTCPConnectWaitTime; // set reconnect timer
Timer.Num[TIMERLOGALIVE] = -1;
}
}
tmo=2000;
NutTcpSetSockOpt(sock, SO_RCVTIMEO, &tmo,
sizeof(tmo));
}
I have about 38K free heap, so I think not that's the problem. I have no
problems with app level reconnects, this code will deal with that.
As I cannot post pictures, I put a retransmission debug to here:
http://www.esh.hu/nut.jpg ( the blue is the nut board ) In this case, the
nut board doesn't reconnect, becouse after a couple of retransmission it
gets answered.
In the case of error, the situation is that, after the retransmission
(sending started by the SkySendStatus(TCP_PACKET_STATUS_NEW,
GetRandomNum()); ), the server not send answer (PUSH ACK) back, but it send
a FIN tcp packet, and this is why nut abort the socket.
And after that,
rc = NutTcpReceive(sock, receivebuff, sizeof(receivebuff)); // see if we get
something
Return -1, what is ok, becouse the socket died.
I'll make some debug picture about the disconnect evening.
The question is, what can be wrong, as a very similar logic used ethernut
based GPRS board can keep alive 2 days on this server, With the same code in
local net this board can keep alive weeks, but on an adsl line this happens.
Regards,
Andras
-----Original Message-----
From: en-nut-discussion-bounces at egnite.de
[mailto:en-nut-discussion-bounces at egnite.de] On Behalf Of Harald Kipp
Sent: Tuesday, April 08, 2008 1:22 PM
To: Ethernut User Chat (English)
Subject: Re: [En-Nut-Discussion] unreliable tcp
Hi András,
Szemző András wrote:
>
> I tried to use Commview to capture packets, and I see the sam7x
> doesn't get answer within a time. Normally it sends a PUSH ACK, get
> answer PUSH ACK, and send ACK. I see when there are problems, it tries
> to send PUSH ACK with retransmission, and if it gets answer it doesn't
reconnect.
> My app's loop send a packet if it needs, than read back the answer,
> and with 100ms RECV timeout tries to read to see, if he got some
> incoming
packets.
> I see when there are reconnects, the NutTcpReceive() will return with
> -1, wich means socket is disconnected. (sock->so_state will be 0 in
> this case)
The reconnect _must_ have been initiated by your application. The stack by
itself will not reconnect. Probably because of the -1 returned by
NutTcpReceive().
NutTcpReceive will return -1 only in case of fatal errors (e.g. out of
memory) or if it receives a connection close (FIN) or connection reject
(RST).
What does NutHeapAvailable() return?
Harald
_______________________________________________
http://lists.egnite.de/mailman/listinfo/en-nut-discussion
_____________ NOD32 3009 (20080408) Információ _____________
Az üzenetet a NOD32 antivirus system megvizsgálta.
http://www.nod32.hu
----
Kometa 99 Élelmiszeripari ZRt.
Székhely: 7400 Kaposvár, Pécsi u. 67-69., Levelezési cím: 7401 Kaposvár, Pf.
58
Cégszám: Somogy Megyei Bíróság mint Cégbíróság 14-10-300239
Adószám: 13749619-2-44
Telefon: 82/502-400, Fax: 82/502-415
http://www.kometa.hu
http://www.kemencessult.hu
_______________________________________________
http://lists.egnite.de/mailman/listinfo/en-nut-discussion
_____________ NOD32 3009 (20080408) Információ _____________
Az üzenetet a NOD32 antivirus system megvizsgálta.
http://www.nod32.hu
----
Kometa 99 Élelmiszeripari ZRt.
Székhely: 7400 Kaposvár, Pécsi u. 67-69., Levelezési cím: 7401 Kaposvár, Pf.
58
Cégszám: Somogy Megyei Bíróság mint Cégbíróság 14-10-300239
Adószám: 13749619-2-44
Telefon: 82/502-400, Fax: 82/502-415
http://www.kometa.hu
http://www.kemencessult.hu
_______________________________________________
http://lists.egnite.de/mailman/listinfo/en-nut-discussion
_____________ NOD32 3010 (20080408) Információ _____________
Az üzenetet a NOD32 antivirus system megvizsgálta.
http://www.nod32.hu
----
Kometa 99 Élelmiszeripari ZRt.
Székhely: 7400 Kaposvár, Pécsi u. 67-69., Levelezési cím: 7401 Kaposvár, Pf. 58
Cégszám: Somogy Megyei Bíróság mint Cégbíróság 14-10-300239
Adószám: 13749619-2-44
Telefon: 82/502-400, Fax: 82/502-415
http://www.kometa.hu
http://www.kemencessult.hu
More information about the En-Nut-Discussion
mailing list