Link Quality Monitoring

Guides for Evaluation of Link Quality

pppd bases its evaluation of the link status on the arguments of two associated options, lqrinterval and lqthreshold. The value of the lqrinterval option tells pppd how often to request LQR packets from the peer machine. The default value is every ten seconds, or five per minute, allowing for timing slop. lqthreshold's argument is the minimum acceptable link quality. The default lqthreshold is one out of five packets. With the default values in place, the link will shut down if no LQR packets arrived during the previous minute.

Adjusting LQM Behavior

The default LQM behavior, one successful packet out of five sent every minute, is fairly permissive. pppd is more demanding about line quality if you change the defaults. Either evaluation setting can be changed to make LQM more stringent. For example, you can demand that at least one LQR arrive per minute by specifying

lqrinterval 5 lqthreshold 1/12

Or you can demand that no more than one LQR be lost each minute by specifying

lqrinterval 5 lqthreshold 11/12 

Weighing the Costs of LQM

pppd discovers line failures more quickly if you decrease the lqrinterval because LQR's will arrive more frequently. However, sending more LQR's increases monitoring traffic, slowing down the transfer of user data. So you should remember to also consider raising the lqthreshold. If the lqthreshold is 'lqthreshold 5/6', no more than one LQR can be dropped per minute. If two LQRs in a row are dropped, pppd shuts down the line.

LQM Response to Link Failures

LQM is particularly useful for detecting and responding to total link failures. The response can include failover protection which switches the connection to an available dial-up modem. Most total failures, for example, a backhoe digging through the telephone company's cable, cause the connection's CSU/DSU or modem to deassert the Carrier Detect signal. pppd observes this as a hangup event.

Failure Without Disconnection

But some failure modes, like misconfigured flow control and over-reliance on in-band XON/XOFF flow control, leave modems connected even though the PPP peers are unable to communicate. In this situation, the peers observe an LQM failure and take appropriate action, usually disconnecting and redialing. LQM is also useful if an intercontinental telephone connection is of such poor quality that significant numbers of packets are damaged in transit. Again, the PPP implementations can decide, based on LQM measurements, to hang up and redial in hopes of getting a better connection. This is a very rare occurrence if your modems provide V. 42 or MNP 4 error correction, but it does occasionally happen. Whatever the cause, the disconnection and redial operation can happen without user intervention or application awareness. This is because even if PPP frames are damaged or lost, the upper protocol layers arrange for retransmission as needed to provide the user with a complete data stream. The user simply experiences a pause while the modems reestablish the connection.

Peer Refusal to Comply with LQM Request

If, during LCP option negotiation, the peer refuses to send Link Quality Reports, pppd instead begins sending LCP Echo-Request messages at the requested lqrinterval and use the arriving LCP Echo-Response messages to make the link quality decision. If the peer does not correctly use a LCP Configure-Reject message to tell PPP to switch to LCP Echo-Requests, PPP can be given either the echolqm argument, to dispense with the negotiation phase and begin directly with Echo-Requests, or the nolqm argument, to disable link quality monitoring completely. At pppd's debugging verbosity level 4, the log file receives summary messages like this:

5/7-13:45:27-1514 LQM: Pkt: 1/1 Oct: 53/53 LQRs: 5/5

This means that, during the last testing interval, this system transmitted one packet and received one packet. Fifty-three octets crossed the link in each direction. And this system has received responses to all five of the most recent five Link Quality Reports it sent. The LQR packet is 36 octets long, and the default lqrinterval of ten seconds will cause the additional traffic to be unnoticed on most connections. However, if the application is very sensitive to speed and requires absolutely every bit of available line bandwidth, use the nolqm argument to disable LQM.