bobkil@ibmpcug.co.uk (Bob Kilgore) (03/12/91)
SECOND POSTING; FIRST WAS INCOMPLETE; SORRY! Keith Dixon of BSO/Automation Technology Breda B.V., The Netherlands Writes: >I am having problems in a large but not excessive Ethernet consisting of a >variety of cabling systems from twisted pair to optical, and equipment >from UB, DEC and BICC. Protocols include TCP/IP and DECnet. > >VAX based terminal servers with CMC ENP55 Ethernet cards seem to be >resetting as queues overflow. The net is not particulary heavily loaded, >but when we did a stress test on the network, we noticed a number of 'late >collisions'. > >If everything is OK, late collisions shouldn't happen. I have never seen >this happen before, but I have never looked for them. We seem to have >about 1 error in 10,000 packets, even if we strip the network to a >minimum. >1) Is this a high level, or is it acceptable ? >2) Has anyone had similar experience, especially with this card ? >3) Does anyone have any idea how a card will (in general) react to > this (illegal) condition. >4) Any advice as to how I should track down the cause, rather than > trying to attack the symptoms. > >Keith Dixon. > Keith; A 'late' collision is, generally, caused by a Node failing to detect the fact that activity is taking place on the media. Since most nodes transmit only rarely, it would cause a net. fault that is dependent on its workload, and random in time. The only cure is to find the Node and replace the offending device. It could be the transceiver, but more likely the controller card. A late collision should not cause any harm to the network, and locating it is difficult and time consuming. You will have to decide if the incident rate is to high. Your level of late collisions indicate to me that the offending device is a user node, not a server. As to your queue overflow problem, I seem to remember that the default setting in the VAX was to low for a mixed protocol network. For instance if you used DEC-net only, then the queue buffer assignment was good, however if you mixed in some TCP/IP and IPX then the VAX queues began to overflow and need to be increased. I seem to remember a posting to that effect. Hope this helps a bit; Best Regards; BOB bobkil@ibmpcug.co.uk -- Automatic Disclaimer: The views expressed above are those of the author alone and may not represent the views of the IBM PC User Group. --