100G, Scale and visibility

Let’s consider two important challenges: Growing scale, but also gaining back visibility.


Growing bandwidth, scale, and port density in an economical manner seems an impossible challenge. Traditionally increased  bandwidth and throughput came at a portional cost to a vendor’s hardware innovation. The adoption of merchant silicon has successfully challenged that perception within the confines  of the data center. Outside of the data center, while merchant silicon is offering an ever increasing richer features set in a faster denser form factor, the improvements have lagged  behind the requirements of a true routing platform, in terms of logic and table size which has been restricted due to the limited space on the silicon die.


To achieve scale Operators have been historically tied to the bandwagon of hardware-defined platforms, where innovation is only achieved at a vendor’s cost point. This approach is now being disrupted by the latest evolution of merchant silicon, which now blurs the line between switching  and routing logic and software innovation which can  extend the default hardware capabilities of the merchant silicon. This leads to a full Internet table being held in  merchant silicon hardware at a quarter of the cost point of a traditional hardware vendors routing platform.


Previous generations of merchant silicon, table scale has been the defining line preventing merchant silicon switches delivering routing functionality at commoditised prices. With latest merchant silicon this is no longer the case; welcome to 100G routers at $3000 per 100G port. They might not fit every needs as the switch vendor might not support your favourite FastReroute or MPLS L3VPN feature yet, but it does disrupt the traditional market of big expensive routers that are big and very expensive.


With increasing scale and performance in the network, the visibility into the network has deteriorated because monitoring technologies and tools haven’t improved or kept apace with the change. In an environment where growth in bandwidth and density is continuous and the effects of  network blips however small are having an ever greater impact, the appetite for better visibility model has been greater.


Streaming of network states comes to the rescue, illustrated by OpenConfig (small but standard scope) or other wider scope implementation (all network states - vendor-defined but open/standard access). Streaming in monitoring is a paradigm change from polling, for efficiency and scale. Imagine one thing you could never do before: receiving all the prefix changes in all the routing tables, ARP resolution, MAC addresses, CPU/temp/fans, optical measurements, logs, counters, etc… all being streamed as changes occur:


Without missing a single change

Without having to ever interrogate the devices


The whole network states can be viewed, coalesced, retrieved from an hour ago, yesterday, or last season. A device worth of streaming represents in the order of 100MB/day, making it a goldmine of information for troubleshooting, analysis, finding anomalies or trends.

OpenConfig is the logical choice for standard-defined data streaming, but much more is also available that can be retrieved equally easily from network vendors’ state streaming.


Poor or expensive scale, and lack of visibility are not fateful, software innovation brings cost-effective scaling to merchant silicon and modern telemetry for complete visibility into the networks.