At the Intel Developer Forum being held in San Francisco, Ericsson is showcasing their brand new Hyperscale Data Systems (HDS8000) which is based on Intel’s Rack Scale Architecture and Ericsson’s expertise in optics. As part of this revolutionary hyperscale offering, Pluribus was selected to power the networking fabric and towards that end, we announced Netvisor 2.3 with features to support the HDS8000’s demanding switching, network virtualization, and Analytics requirements. The Ericsson HDS8000 is truly a next generation architecture, so it was no surprise that Ericsson selected Pluribus to power their networking layer.
Before discussing Pluribus’ Netvisor technology, let me explain the need for a completely new rack architecture.
Intel Rack Scale Architecture (RSA) to drive the next generation racks
Intel RSA is the first major revolution in data center design being seen for more than a decade. The last major revolution happened in the early 2000s when Intel created the core architecture for its CPU which resulted in “Nehlam” processors with multi-core, hyper-threading and H/W based virtualization that created the server virtualization and the ubiquitous dual-socket commodity servers that run today’s data centers. Integrated PCI-express controller on CPU and copper-based 1g and later 10g connectivity made this commodity architecture very resilient and price competitive. Since then, we have seen incremental improvements in cores, threads, and frequency but no major shift to the fundamental design of a data center. Increasing CPU frequency or the number of cores and chips alone is not sufficient to address next-generation needs, since I/O has started to become the bottleneck. On the I/O front, copper interfaces are very cost-efficient but not a good medium for bandwidth beyond 10 gig. This applies to all I/O, including Ethernet, PCI-Express, and SATA cables. Intel proposed their Rack Scale Architecture (RSA) to disaggregate the processing “CPU” complex from its associated I/O complex by introducing the use of an Optical Interconnect to overcome the speed limitations, allow upgrade of CPU complex separately from I/O and storage and ultimately increase the rack density to meet the need for next-generation data centers.
Ericsson Hyperscale Data Systems provide Optical based I/O for scale
Ericsson’s Hyperscale Data Systems (HDS) is not just another aggregated attempt for building public and private clouds. They used their long Intellectual Property and experience with Optical Interfacingthat had allowed them to dominate the mobile and Telco markets for decades and added “Silicon Photonics” to do optical switching. With cost in mind, they got rid of the optical transceivers that often cost more than the switches themselves which also gives the perception that optical infrastructure is very expensive. In a nutshell, for RSA to succeed, we needed an inexpensive way to convert a digital signal into an optical one and back again. Today it is achieved by using optical transceivers and ancillary chips to boost the signal strength (in the case of copper). In simple terms, using Silicon Photonics, Ericsson has added the ability to convert digital signals to optical signals in the PCB itself eliminating the need for expensive transceivers and optical cables to plug in directly.
Optical cables without the transceiver are significantly lower-cost then their copper equivalents and have very little signal attenuation. By using optics as Ericsson is doing, the payload data can be carried much longer distance without the need for signal boosters. For comparison, 10gig signals can run a couple of feet in copper and is reduced to less than a foot when considering 25/40/100 gig. With optics, the same 25/40100 gig signals can traverse 1000s of feet…creating the revolution we are speaking of. The RSA disaggregation model now allows for CPU/memory board connected to Ethernet boards and storage boards over chassis-based optical backplane where both Ethernet and SATA are carried over optical cables (and PCI-Express coming in near future for even more switch disaggregation). Think bigger Lego-style building blocks that allow upgrading to any system component independently, dramatically better scale and performance of 25/40/100 gig at commodity pricing. It’s no wonder that Ericsson based its core hyperscale offering upon Intel’s RSA architecture as it is the only tangible invention in more than a decade to enable true mega-center hyperscale and associated rack densities that will challenge the imagination.
Netvisor, the Operating Systems for Hyper-Converged devices powers HDS Networking
Netvisor is a bare metal distributed switch OS that Ericsson has selected to drive the switches in their next generation racks. The development work between Ericsson had Pluribus has been going on for close to a year and final production readiness steps will start in the near future.
- Netvisor’s Cluster-Fabric has a distributed control plane to securely and seamlessly aggregate all of the switching components and allow centralized management of the resulting ‘fabric’. Netvisor also includes a robust set of RESTful/Java/C/etc APIs to integrated with Ericsson’s HDS command center. Dual redundant in-band and out-of-band management of the control plane running over TLS provides very secure and the highly available environment required for mission-critical environments and hyperscaled cloud systems. Pluribus’ fabric wide virtual port table “vport” provides connectivity for Virtual Machines independent of the Layer2, Layer3 or Overlay based connectivity.
- Netvisor’s Virtual Networks “VNET” allows micro-segmentation of the network and offers per VNET neutron plugin/APIs. This allows the network itself to be virtualized and integrated with VNET cloud managers (Openstack, vSphere, Xen, etc). Netvisior’s virtualization allows virtual networks to be created regardless of the hardware layer itself. VNETs can span any number of switching devices that take on the logical appearance of a single-multi-port device and management domain. And these domains are configurable by interacting with the fabric itself, from any point within the fabric, and all changes are propagated instantly throughout the network.
- Netvisor’s Integrated Application Analytics allows operators or cloud managers to have a full view of VMs and applications without needing to deploy costly external TAP hardware devices. The costs associated with this older TAP hardware approach has been prohibitive for wide-scale application, so sampling or representational placements have historically been chosen to keep costs low. But along with saving costs, the visibility to data flow information was dramatically reduced. Netvisor’s Integrated Application Analytics makes complete visibility a reality through its virtual TAP architecture. Ericsson uses Netvisor’s REST and C APIs to integrate application analytics in their “Data Lake” along with their other analytics to give customers full view of the data center and meet their SLA, compliance and security needs.
Open Netvisor 2.3 extends server style management to today’s advanced networking switches while combining it with the programmability and automation of SDN. Networks Operations teams will feel at home managing a Netvisor-based fabric (right down to our very familiar CLI), and the DevOps teams will feel empowered to interact with the network in ways they had thought would never be possible. It’s easy to see why Ericsson chose Netvisor to power its next generation flagship hyperscale offering, it’s HDS8000. Ask us to show you how we can do it for you!