Maxing out connection causes disconnects

When running torrents overnight, this quite often maxes out my connection. After a while running at, or close to maximum speed (about 1.4Mb/sec) I get a disconnection. The firewall (IPCop) immediately reconnects and the transfers take a few minutes to get up to speed again. In the two hours between midnight and 2am this morning I had 12 individual disconnects. When running regular browsing the router can go for weeks without reconnecting.

The log from the RED NIC on the firewall shows:

00:17:01 pppd[26329] rcvd [LCP TermReq id=0x3]
00:17:01 pppd[26329] LCP terminated by peer
00:17:01 pppd[26329] Script /etc/ppp/ip-down started (pid 5472)
00:17:01 pppd[26329] sent [LCP TermAck id=0x3]
00:17:03 pppoe[26331] Session 2537 terminated -- received PADT from peer
00:17:03 pppoe[26331] Sent PADT
00:17:03 pppd[26329] Script /usr/sbin/pppoe -p /var/run/pppoe.pid -I eth1 -T 80 -U -m 1412 finished ( pid 26331), status = 0x0
00:17:03 pppd[26329] Modem hangup
00:17:03 pppd[26329] Connection terminated.
00:17:03 pppd[26329] Connect time 2436.5 minutes.
00:17:03 pppd[26329] Sent 205256909 bytes, received 3346672669 bytes.
00:17:03 pppd[26329] Waiting for 1 child processes...
00:17:03 pppd[26329] script /etc/ppp/ip-down, pid 5472
00:17:04 pppd[26329] Script /etc/ppp/ip-down finished (pid 5472), status = 0x0
00:17:04 pppd[26329] Exit.
00:17:04 red: Connectioncheck will connect again

The reconnects do not seem to affect my sync speed, in fact in the past week or so my connect speed has improved from 15320 to 17536 and my profile has gone from 13000 to 15000. For the record I got the same disconnects at the lower speeds too.

Unfortunately I can't pull any line stats from the modem in bridging mode.

My hardware setup is:

Netgear DM111P ADSL2+ ethernet modem (in bridging mode) connected to IPCop 1.4.21

Anyone shed some light on this for me?

Thanks,
TH.

Edited by deleted (Fri 21-Jan-11 10:30:51)

The number of packets the DM111 is handling is probably causing its CPU to crash

In reply to a post by Trojan_Hussar:

00:17:01 pppd[26329] LCP terminated by peer

This shows that the PPP session was dropped, but not nessarily sync. You would need to check your modem to see if sync was lost as well.

In reply to a post by Trojan_Hussar:

Unfortunately I can't pull any line stats from the modem in bridging mode.

You should be able to, but you need to make a change either to the RED interface address on IPC or the LAN address of the modem so you can communicate outside of the PPP session.

In reply to a post by Trojan_Hussar:

Anyone shed some light on this for me?

If it's just the PPP session dropping (and sync is maintained), it could be for several reasons. e.g. ISP problems, PEW, IPC choking with the number of connections etc.

I'm currently working on the theory that by maxing out the connection I'm preventing LCP pings being answered and once the router fails to receive acks for three LCP requests it assumes there is a I/O error and issues a PADT packet.

I've altered the "lcp-echo-failure" from 3 to 6 in the PPPoE section in /etc/rc.d/rc.red on IPCop, if this doesn't work I may raise the lcp echo interval from 20 seconds. Will report back once I've run some more heavy overnight transfers.

Thanks,
TH.

Can you prioritise outgoing acks via QOS?

In reply to a post by BatBoy:

Can you prioritise outgoing acks via QOS?

I'm assuming it's the returning acks failing to get through. I've increased the number of ack failures to 6 before it issues a PADT, if this fails I can also increase the time between LCP pings (it's currently 20 seconds).

Theoretically if the bandwidth was totally preventing the LCP acks we should see a disconnect every 40 seconds (3 consecutive failed lcp pings @ 20 second intervals) the log is showing some periods where I'm getting disconnects every 2-3 minutes and others where it's disconnecting every ten minutes or more. Hopefully raising the failure threshold to 6 should reduce the dropout rate.

Once I've run some transfers tonight I'll post the result here.

Thanks,
TH.

Yeah, Mixt wrote a document about prioritising outgoing acks using ipchains iirc.