LANCP unavailable user buffers

[Gerald Kerner]

I am copying via DECnet over a DS3 between ES40's both at 7.3-2, and the link speed seems to vary greatly. In LANCP the only 'error' listed is Unavailable User Buffers at both ends. Can I make more user buffers? How? Thanks Jerry

[labadie_1]

Do you use Decnet IV or or Decnet OSI ?

For Decnet Phase IV, can you post

$ mc ncp sh k line counters
$ mc ncp sh 'remote' counters
$ mc ncp sh exe counters

(remote being your remote node)

and the same on the remote node (replacing remote by your first node)

May be there is some routing involved, do the Decnet addresses are in the same area (for example, 20.60 and 20.61) ?

[Volker Halle]

Jerry,

welcome to the OpenVMS ITRC forum.

With ANA/SYS and SDA> SHOW LAN/COUNTERS you may be able to see, which protocol is incuring the Unavaliable User Buffer events and when the last UUB event has happened. Does this information correlate with your DECnet copies ?

Please note that whenever DECnet packets get lost, the retransmission time is typically significant and is often interpreted as 'bad performance'.

Volker.

[Paul Beaudoin]

Gerald,

Short answer: I don't believe you can add more buffers to DECnet IV (not sure about V). As Volker points out, either version of DECnet will recover the lost buffers and if there are not LOTS (an undefined largeish number) of then, this should not cause a problem. Unavailable User Buffers are just that: It is the user defined space where the driver (probably Ethernet in this case) copies received data to. The number of buffers is defined to the driver when the protocol starts and when you have control of the code, or create an appropriate interface, you can control this number at startup. It has been some years sine I looked at this and certainly then (c. 7 years) , there was no solution. Nor was there a practical problem - more an irritating anomoly.
Regards
Paul

[Richard Stockdale]

You can look in SDA SHOW LAN/DEV=DECnet to see the "Rcv buffers to save". If a receive packet comes in and there is no available IRP to deliver the receive to the application, the LAN driver will queue up to this number of buffers internally.

For DECnet-IV, the NCP SET LINE dev RECEIVE n will affect this number, and in your case, should reduce the number of packets lost. You might have to adjust the pipeline values as well to stop each side from sending more data than the other side can receive without loss.

[Gerald Kerner]

Thank you for your responses; sorry for the delay in my answer but I was unable to login. After no VMS issue was found the network was reinvestigated and 100% utilization turned up. No VMS problems. No data was lost.