Blog

Networking and the White Box Solutions

The recently published AT&T white paper “Towards an Open, Disaggregated Network Operating System” shows the company’s willingness to discuss and actively promote the creation of the industry-standard open NOS, intended for adoption by the AT&T network. Installed on white box network devices, it might potentially replace over 100,000 proprietary AT&T routers in the future. The dNOS represents a call for technically feasible, cost-effective, vendor and hardware independent solution running on white box (merchant silicon based) networking devices. The concept has potential to change the entire networking industry in the future, pushing network vendors to move from combination of proprietary software bounded with purpose-built hardware to a disaggregated model with software separated from underlying hardware.

The dNOS platform proposed by AT&T running on white box equipment relies on standard interfaces and Application Programming Interfaces (APIs) that allow customization and enable modular design. The dNOS platform is intended to cover all size, power, functionality and security requirements of a specific deployment, while enabling network operators to mix applications from multiple vendors. This would add necessary flexibility and scalability to a network design with the plug-n-play hardware while decreasing the costs typically associated with the use of standard hardware.

As we know, traditional proprietary network equipments contain thousands of custom application-specific integrated circuit (ASIC) chipsets in order to meet the specific performance requirements. For instance, Cisco Nexus 7000 uses Cisco-designed proprietary ASICs. Custom ASIC or proprietary silicon is the specialized hardware designed for routing and forwarding packets at very high speeds. The development of custom ASIC chipsets is however a resource-intensive and time consuming process [1]. Moreover, each network device family often needs its own custom ASIC, which also increases the cost of the custom silicon development. Therefore, specification of custom ASICs and their corresponding Software Development Kit (SDK) has been kept in secret by ASIC vendors.

Thankfully, the growth of merchant silicon in the network market and open SDKs enable the white box network vendors to use merchant (off the shell) silicon in their equipments. The white box vendors do not design their own chips but use the merchant chips instead. For instance, Arista Networks, one of the key players in the Ethernet switch market uses “SDN” (highly programmable) XPliant chips from Cavium in their 7160 family switches. Therefore, it might be considered as a fact that generic merchant chips fulfill current capability and throughput requirements.

Some network vendors such as Juniper, Dell, Avaya or Arista have already adapted to the global demands of network operators for dNOS, bringing their own solutions to the network market. For instance, Juniper has developed disaggregated version of Junos OS which can run not only on Juniper hardware but also on white box or bare-metal servers. It is modular, based on Linux and enables customers to manage the system as a server. Installation of third-party applications such as tcpdump, Quagga is also supported on a virtual machine or inside the containers. Applications can be installed and upgraded from standard Linux repositories, independently of Junos OS release.

Facebook has its own data center top-of-rack network switch Wedge, running Linux-based operating system, Facebook Open Switching System (FBOSS). FBOSS consists of FBOSS agent that programs and control a single merchant silicon from Broadcom, Trident II for high-speed forwarding. FBOSS agent is a central point of FBOSS and its function is to program L2/L3 tables within the Broadcom ASIC, parse packets and provide the low-level control of packets for host and neighbor learning. (ARP, IPv6 NDP, DHCPv4/v6 relay, LLDP) [2]. Rather than being a full NOS, FBOSS should be treated as a set of applications. For instance, a low-level application such as FBOSS agent is responsible for ASICS programming. Facebook uses Broadcom open library of network switch OpenNSL for this purpose. The control applications implement routing protocols such as BGP. And finally, automation applications are destined for configuration, monitoring, and troubleshooting of the switch. The configuration of the FBOSS is done via API rather than CLI. Therefore, configuration changes can be applied to thousands of FBOSS switches at once.

We have already mentioned some benefits of the white box network equipments. The most significant benefits are lower hardware cost, choice of software platforms, faster adoption of technologies, no vendor lock-in, lower cost of maintenance and lower operating costs (licensing). However, as hardware and software are disaggregated, two different vendors – the hardware manufacturer and the operating system vendor are involved in the processes of making the equipment. Sometimes, it is hard to determine whether the issue is software or hardware based. Therefore, support might be a concern as it is not clear who supports the white box equipment in this case –  is it NOS or hardware vendor? Support of  black boxes is more straightforward as the manufacturer typically supports a device as a unit, including  both OS and hardware components.  


Conclusion

:
To really “open” the networks in the future, the white-box deployments should meet several criteria. First, the hardware with an open API is a must and it should be offered for lower prices while providing similar or better throughput than its counterparts from the proprietary world. Secondly, offered as the bare-metal boxes with minimal boot-loader, white boxes should be able to run any Network Operating System according to the customer’s’ needs. Compatibility between software and hardware can be assured by the white box vendors provided certification program.  And finally, the white box vendors support should be equal to the support offered by traditional network vendors in terms of quality and availability.