MSCP and multiple Ethernet connections

[Ronald Sisson]

If your OpenVMS systems have multiple NICs, will the MSCP services use more than one NIC? Can you configure which NIC(s) is/are used?

[Volker Halle]

Ronald,

MSCP works as a SYSAP (System Application) above the SCS (System Communication Services) layer. Communication between nodes in a cluster is provided via the Virtual Circuit (VC), which can be multiplexed over various Channels. A Channel is a communication path between individual NICs of clustered systems.

You can enable and disable SCS (cluster protocol) on each NIC (= LAN DEVICE) in a system.

With SCACP, you can assign priorities at the LAN device and CHANNEL level to control SCS traffic. By default, SCS will be enabled on all LAN devices and will form channels between the LAN devices on the local node and all reachable LAN devices on the remote nodes.

You can therefore influence which channels are preferred for the virtual circuit, but you cannot control the traffic at the SYSAP layer.

Volker.

[Hoff]

What problem(s) might you be seeking to resolve with your question? Could you provide some background?

[Ronald Sisson]

We have a production cluster with a dual-fabric SAN (NSPOF configuration) providing access to two EVA5000 arrays. The production disks are volume shadowed across the two EVAs. We are configuring a disaster recovery site with standby systems (not booted). We will be relocating one of the EVAs to the standby site and extending the two fabrics, so that the second EVA is still providing active shadow set members.

In the scenario where one of the primary hosts has failed, the equivalent standby system will be booted into the cluster. At this point, I want to ensure that there are MSCP paths to the disks at both sites, in case both fibre paths between the sites are lost.

Each system has 3 NICs, each on a separate VLAN. Two of the NICs have SCS enabled, while the other is used for DECnet and IP only. This is to provide two cluster communications paths, and keep cluster traffic separate from DECnet and IP traffic.

The previous response seems to indicate that both of the SCS paths will be available for MSCP serving. Will both be active automatically? Will one be chosen arbitrarily, and if so, will it fail over if one of the two VLANs is lost?

[Volker Halle · Accepted solution]

Ronald,

SCS will detect all possible communication pathes between the LAN devices of all systems in the cluster. It will form a channel between each pair of reachable LAN devices. As long a one channel exists between any two systems, the VC will work and be able to transprot data for all SYSAPs.

Have a look at it with:

$ SET TERM/WI=132
$ MC SCACP
SCACP> SHOW LAN ! show local LAN adapters
SCACP> SHOW CHAN ! show all channels
SCACP> SHOW VC ! show virtual circuits

Volker.

[Hoff]

Rather than looking at controlling SCS via the NICs (save for those cases where there are widely disparate speeds and feeds among the NICs present; then see SCACP, et al), I'd concentrate on the processes and procedures around the detection and remediation of failed SCS paths. This is the case that can sneak up on you, not typically anything around not finding new paths -- that's automatic. It sucks big-time when the cluster has calmly and transparently failed over to the last available path, and then that final last available path vaporizes.

There are similar considerations around managing dual-redundant storage controllers. When one fails, there is some squawking -- but OpenVMS will select and will continue to operate on the remaining controller.

Stephen Hoffman
HoffmanLabs

[Ronald Sisson]

Thanks for the answers!

[Ian Miller.]

while you are looking at this then look at implementing
LAVC$FAILURE_ANALYSIS

(see cluster manual)