Best Practices for Scaling DDC Clusters

What’s the problem with networking?

Designing a network is not a trivial task. You are always fed with sub-optimal data as for what’s still to come. You are expected to build something that both fits current needs and is ready for whatever happens next. It’s kind of an ungrateful task that goes unnoted even when done right.

A common approach is to oversubscribe and position a larger device than needed, keeping in underutilized most of its life span. This is costly in terms of both CapEx and OpEx. Sometimes a device runs its entire lifespan at under 50% utilization, but at least this avoids the dreaded hardware (HW) replacement alternative.

Beyond that low utilization issue, every site can be occupied by a different-sized network element that hosts different line cards and requires a different spare parts kit. When HW devices need extra capacity, more line cards are added until they are full. The next growth step is to replace the device with a larger one – hopefully from a similar series so that feature parity is enabled, but this too isn’t guaranteed.

Few points regarding Distributed Disaggregated Chassis (DDC)

A DDC is comprised of the same components as a traditional chassis except for the chassis enclosure itself. This means that the number of line cards you add to an already deployed router is not dependent on the number of free slots in the chassis selected for that location; rather, its dependent on the size definition applied to the DDC cluster upon its deployment.

DriveNets Network Cloud runs on standard networking white boxes supplied by a variety of original design manufacturers (ODMs). Based on merchant silicon (OCP DDC-compliant), these networking white boxes comprise just two building blocks – Network Cloud Packet Forwarder (NCP) and Network Cloud Fabric (NCF).

[I copied this new paragraph above from DriveNets site; this sort of brief intro to NCP and NCF seem necessary if we are going into more detail about them below]

When selecting the size of the cluster, the consumed space and deployed devices are reduced to the NCF components only, while NCP components can be added at later stages upon growth of the network. Even when the defined cluster is exhausted, redefining the cluster as a larger-sized router is a valid option; the boxes installed will maintain their roles so there will be no need for any forklift or insertion of new device types to the network. Even starting the router as a standalone single device and growing it into a cluster is a possibility. DDC makes scaling a real option.

Tips and pointers for DriveNets Network Cloud deployments

1. Pre-provision the configuration of an NCP before you physically insert it into the network
2. Create a golden config template per service for smooth deployment and faster “time to service”
3. Scaling can also mean scaling down, so Network Cloud Elements (NCE) can be added or removed from the network
4. Set the added/removed NCE to admin-state disable before adding/removing it from the network
5. Don’t exaggerate the defined cluster size; while this is a unique capability, there is still some cost involved with overshooting the scale of the router
6. Use DriveNets Network Orchestrator (DNOR) to admit the new device to the DDC cluster
7. Scaling up the entire cluster requires rewiring of fabric cabling and is traffic disruptive – allocate a maintenance window before taking that option

X factor for Networking

Scale up or down while staying with the same hardware type. Redefine the router as a larger one and simply keep growing it by adding more similar type white boxes. These are attributes that don’t really have a match in the legacy world of monolithic routers. It is not a case of doing it better but more about having this capability at all.

Still, this blog post is about doing it right. “With great power comes great responsibility,” says our “marvelous” neighborhood friend – so stay focused on the tips and pointers.

How is Distributed Disaggregated Chassis (DDC) a game changer for networking?

DDC is a game changer when it comes to rescaling an existing router. This capability sets a new standard for network planning and maintenance.

Also, since DDC is an OCP standard, you can get the same device type from different manufacturers – removing vendor lock-in and allowing freedom for all.