|
HP OpenVMS Systemsask the wizard |
|
The Question is: I have 2 Alpha E45'S in an openvms cluster. Each node has 6 ethernet interfaces. 2 x de602's with 2 ports each 2 x 1Gb ethernet adapters. There is only one LAN, but I still want to configure for availaibilty and throughput. My idea is to run SCS traffic primarily on one of the 2 1Gb ports and maybe 1 each of the de602 ports, just by linking ports via ethernet cable, no decnet plus nor tcpip on those ports. SCS operates at a lower network layer than decnet &ip. Also I would like to isolate decnet plus to one ethernet port (just used for copies and monitoring to other decnet nodes on network). In DECnet plus, what commands do I need for this? The rest would probably run tcpip. I would configure that to be node1-1 10.xx.xx.100, node1-2 10.xx.xx.101, node2-1 10.xx.xx.102 and node2-2 10.xx.xx.103. Would this work, is there a better way of configuring it? thanks Nigel The Answer is : Please allow OpenVMS to configure and to then operate the available NIC(s) and LAN segment(s) for networking and for clustering. For best overall availability, please keep the configurations simple and consistent. Please consider where the typical outages and typical failures actually occur, and the typical results of one (or more) of these failures. If a PCI controller fails, OpenVMS will usually fail and (assuming a hard error) will reboot without reconfiguring the (failed) controller into the running system. But quite commonly, human error is to blame for most of the outages -- either mistakes at the system management level, or by simply and manually disconnecting the LAN segment(s) or the system power or similar, or by deleting one or more critical data files. While obviously quite tempting, most attempts to partition clustering and networking traffic and networking protocols actually tend to reduce the overall configuration availability. This reduction in addition to the resulting configuration complexities, and increased management difficulties. In turn, these increasing complexities tend to increase the likelyhood of operator-induced (human) errors, and the resulting outages -- and increased times for restoration. Data BACKUP operations and (tested) data recovery plans are particularly vital. Measuring and planning for system growth is also important. Forcing all IP routing over a particular LAN segment (for instance) is rather less critical, and -- if the LAN path fails -- more problematic. Put another way, plan for the common errors.
|