[Info-vax] PEA0: Errors

[JF Mezei]

Christoph Gartmann wrote:

> Both, the VMS boxes and the switch are set to fixed rates. The Alphas have
> 100MB/FD while the VAXen use 10MB/HD. One strange thing is that one Alpha
> needs "auto negotiation" on the switch, otherwise the connection is flaky.

Try putting the switch to autonegotiate on the VAX links as well.


> This is clear. But what does this HP switch mean when it states "drops"?

Not sure of the HP terminology. But I would assume that it would refer
to the loss of carrier. When you plug the cable into the switch, it
takes a few seconds for it to see that there is an active ethernet
cable, negotiate the features ad detect loops.

loss of carrier would be indicative of a bad cable connection.

On Cisco switches, you can define a carrier delay, the moral equivalent
of RECNXINTERVAL (how long carrier can be gone before you declare the
link to be down). But raising that value, you may be able to ride out
temporary glitches on the link.


another possibility for the word "drop" would refer to dropped packets.
If you have traffic flowing at 100mbps from many machines towards the
poor vax at 10mbps, the switch's buffers may overflow, forcing it to
drop packets because it just can't funnel many 100mbps flows into a
single 10Mbps link without losing packets.


If the VAX is meant to talk only within the cluster, you could reduce
the problem with the use of VLANs.

your cluster nodes are all in VLAN 10.  The other nodes at in VLAN 20.
All nodes in the cluster but the VAX are in both vlan10 and vlan 20.

This way, the VAX would not see any of the vlan20 traffic, it would only
see trafffic from VLAN10 so it would not be affected by a malfunctioning
 gizmo in vlan20 (but your other cluster nodes would be, but tehy can
handle the load at 100mbps).