Decnet Phase IV drops adjacency on large file copies

[Roger Grauwmeijer]

We have recently started getting intermitent problems restoring Rdb databases via DECnet. During a restore the remote node suddenly becomes unreachable and the job fails and the adjacency returns about 5 seconds later.
DECnet is stable during normal operations and only occurs on these large restores or large file copies.
There are no errors recorded for the NIC on either node. Is this likely to be some buffering problem we can tune for somewhere?

[David B Sneddon]

Roger,

What are you using to connect the boxes on the
network? Switches, repeaters, hubs?
I have seen similar problems that turned out
to be a faulty DECrepeater port. It worked
OK with normal traffic but whenever the load
increased significantly things started to go
wrong. We tried a different port and that
seemed to fix it. After testing the original
port after hours and confirming that the port
was flakey, we ended up replacing the repeater.

Regards
Dave

[DICTU OpenVMS]

What versions do you use ?

Also, we recently had a similar problem. A copy from one of our nodes to another node (PUT) over decnet wouldn't work, if it was a big file. A copy from the other node to get the file (GET) did work. After lang research, the sysgen parameter RMS_DFNBC seemed to be the problem. The node with the problems had the value 65, other nodes had other values (most default : 8, some 10 or 15 or 20). After changing the value of 65 to 8, the copy of a big file did work. Even the values 64 and 66 did work. This problem accured on a system with VMS7.3-2 with most recent patches. The strangest part is that the copy did work one day and the day after not anymore. Without any changes or reboot at all on the system.

[WilliamSmith11]

There is some time a customer was getting more or less the same problem that you are describing ,
I do a change in the network card :

we note that the network card and the ethernet switch were set at full duplex ,

then we set network card to auto-negotiate,
the problem was solved.

first see if the light in the netword card are at 100mbps. (orange and geen light on)

[Tim Hughes_3]

Generally Iâ ve found its best to ensure everything is configured to 100/full and avoiding autoneg, particularly with older NICs. On VMS you need to set a console variable for this.

Iâ ve also seen issues with the network giving similar problems to yours. Small packets are OK but large packets get dropped. Check the DECNET counters (not the NIC counters). If you also run TCP/IP try pings with large packet sizes > 2000 (eg ping â s 3000 nodexyz). Network folks take ping failures seriously,

[labadie_1]

if you copy from alpha1 to alpha2, what gives, on alpha1

mc ncp show node alpha2 counters

do you see many "response timeouts" ?

and on alph2, the similar command ?

do mc ncp zero k node counters to have "new" statistics