Nexus ACI

Application Centric Infrastructure [ACI] has been described as Cisco's answer to SDN. At its core, it applies security labels to encapsulate packets. Encapsulation done in 2014 required custom silicon and was outside the reach of merchant silicon of the day. It may be possible to do ACI with today's programmable pipeline ASICs -- or maybe not.

The 9300 models appear to be the oldest of the fleet (2014). They are a hybrid design with Broadcom Trident2 forwarding engines and a custom Cisco ALE [Application Leaf Engine] chip to fix the warts. That the hybrid products have been discontinued and Cisco now uses full custom suggests that the original products were a stand in for an ASIC that had fallen behind in development.

In 2016 Cisco introduced Nexus 9200 ACI switches. ACI stands for ACI Spine Engine. The ACI ASICs work stand-alone -- without a Trident-2 forwarding engine.

In March 2016 Cisco added the LSE [Leaf Spine Engine] which is touted as the first 16 nM geometry switch engine. It has a two-slice design with 40 MBytes split two ways. The LSE has Elephant trap auto prioritization which will keep large flows from trashing the performance of small flows.

In March 2017 Cisco added the FX version of this ASIC.

In December 2017 Cisco announced the FX2 which provides netflow collection and export.

In October 2018 Cisco announced the GX which has 80 MB of buffer split across four slices. It is described in BRKDCN-3222.

In June 2020 Cisco announced the GX2 with 100 MB of buffer. There are two ASIC varients: GX2B with 2 slice-pairs and GX2A with 4 slice-pairs. Exactly what makes slice-pair is not clear. Perhaps there are separate slices for input and output. The ASIC has 7 nm fabrication technology and is built out of chiplets rather than an integrated single die.

Microburst monitoring is a feature in ALE, ASE and LSE ASICs. This chapter is from CiscoNexus 9000 SeriesNX-OS Quality of Service Configuration Guide, Release 7.x first published January 2015 and modified September 2018.