The Aurora 750 is based on Barefoot Tofino switching silicon. It has 2x 10G SFP+ ports and 64 x 100GbE QSFP28 interfaces, which can be configured up to 256 x 25G with break-out cables for high-density scenarios.
It is a unique network development platform, designed to bring the twin pillars – performance and programmability – together for the first time in the history of networking. The combination of the Tofino programmable switch chip, the P4 programming language, and Capilano toolset are revolutionary.
Barefoot Tofino™ – the world’s fastest switch silicon, that happens also to be programmable.
P4 – Programming Protocol-independent Packet Processors – www.p4.org exists now as an independent entity to develop a rich open source ecosystem. Unlike human languages, a programming language such as P4 can unambiguously specify the behavior of network forwarding behavior. A programmer can describe the behavior once, then compile the program to run on a variety of different platforms.
Capilano – the compilers and development tools needed to compile and debug programs to run on Tofino.
A ready-to-use NOS, SONiC – a collection of networking software components required to have a fully functional L3 device, is available for data center applications.
- Rapid Prototyping And Deploying New Protocols.
Use intimate local expertise to introduce your own custom protocols to isolate one tenant from another while allowing interconnection of tenants with their applications.
Extend a standard encapsulation protocol to accelerate load-balancing. Implement custom congestion control mechanics, OAM, discovery, and high availability protocols.
- Remove Unused Protocols, Simplify And Streamline The Network.
Precious switch resources are often hard-coded to protocols you don't use. Imagine that you can reduce the size of the L2 forwarding table and re-purpose the memory to increase the L3 IP routing tables instead. With Tofino, for example, a DC can easily increase the capacity of an IP routing table from 300K to 1.2M, allowing them to build even bigger networks and address many more servers. Throwing out unused protocols also means less to go wrong; data-center owners report outages caused by protocols they do not even use but were hard-coded into their switches! With P4, you only include the protocols you need, focusing precious tables as-needed, simplifying the switch and making outages less likely.
- Ensure Complete Visibility Of The Network And How It Process Every Packet.
With P4, Capilano, and Tofino ASIC network operators can quickly add powerful monitoring, analysis, and diagnostics features for themselves, in the field - and our users have already started to do so. One very popular technique, made possible by P4-capable switches, is called “In-band Network Telemetry” (INT). In a nutshell, the network operator decides exactly what information she wants to observe: For example, the precise latency taken by a packet through each switch it is passes through, or the other packets it shared a queue with, the version of the software, the table entries it matched on. Every data packet can be recruited as a probe, without creating any new traffic. Such unprecedented visibility is made possible by placing programmability in the operator’s hands. And of course, a baseline implementation is already available as the open-source “INT.p4” program. Programmers are already looking at how to fully-automate data collection and remediation, as a step towards making self-managed networks.
- Integrating Middlebox Functions Into Every Switch
You can program the features they need directly into your network, eliminating huge numbers of expensive middleboxes. In most cases, the middlebox functions operate much faster than before, because they run on Tofino at full line-rate, rather than on a conventional CPU. Firewalls, intrusion detection systems, address & port translators, traffic de-duplicators, etc. will benefit a lot from this approach.
- Implement Part of Distributed Applications Directly In The Network.
A big data center runs many huge distributed applications; and also has a network with tens of thousands of switches. It is interesting to ask if the switches can accelerate distributed applications, offloading the servers. Recently, researchers demonstrated how the Paxos consensus protocol can be added to the network by implementing a portion of it in P4, and added to switches. This led to many of orders of magnitude acceleration of distributed applications. Other have built new key-value management services directly into the network data plane. Many new fast in-network services to be seamlessly integrated into networks, for free.
- 64x 100/50/40GbE QSFP28 ports in 2 RU
2x 10G SFP+ ports
Up to 256x 25/10G SFP28 port via break-out cables
- 6.4Tbps Barefoot Tofino DFN-T10-064D
- 5000 million packets per second
- Intel® Xeon® Processor D-1527 quad core processor for application deployment
- Up to 16GB of DDR4 memory (8GB default)
- Configurable pipeline latency enabling sub-800 ns operation
Reliable hardware platform
- Redundant 1200W 1+1 power
- Redundant N+1 cooling
Network OS (NOS) options
- Open Network Linux is a Linux distribution for "bare metal" switches, that is, network forwarding devices built from commodity components. ONL uses ONIE to install onto on-board flash memory. Open Network Linux is a part of the Open Compute Project and is a component in a growing collection of open source and commercial projects.
- SONiC is a collection of networking software components required to have a fully functional L3 device. It is designed to meet the requirements of a cloud data center. It is fully open-sourced at OCP.
||US$12,000 for Aurora 750 bare metal switch preloaded with ONIE
||64x 100/50/40GbE QSFP28 ports in 2 RU
Up to 256x 25/10G SFP28 ports via break-out cables
2x 10G SPF+ ports
1x RJ-45 out-of-band (10/100/1000) management
1x RJ-45 console (RS232)
System status LED
PSU1 status LED
PSU2 status LED
||Switching silicon: 6.4Tbps Barefoot Tofino BFN-T10-064Q
Forwarding rate: 9Gpps
Latency: <800 нс (PHY-less)
Packet Buffer: 22MB
Intel® Xeon® Processor D-1527
8GB DDR4 ECC (up to 16GB)
128GB M.2 SSD
||1200W 1+1 RPSU 80+ Platinum:
100V~240V AC / 50~60Hz
Typical/Max power draw: 730/860 Watts (with optics)
||4 N+1 redundant fans
||2U, 600 x 440 x 88 mm
||Operating temperature: 0~45°C
Operating humidity: 20-95% maximum relative humidity (non-condensing)
|EMC and safety
CE Declaration of Conformity
Reduction of Hazardous Substances (RoHS) 6
||Open Network Linux
* QoS - ECN
* QoS - RDMA
* Priority Flow Control
* DHCP Relay Agent
* SONiC to SONiC upgrade
* One Image
* ACL permit/deny
* Tunnel Decap
* Post Speed Setting
* BGP Graceful restart helper
* BGP MP
* Fast Reload
* PFC WD
* MAC Aging
* LACP Fallback
* MTU Setting
* Vlan Trunk
* IPv6 ACL
* BGP/Neighbor-down fib-accelerate
* Port breakout
* Dynamic ACL Upgrade
* SWSS Unit Test Framework (best effort)
* ConfigDB Framework
* Critical Resource Monitoring
* MAC Aging
* IPv6 ACL
* BGP/Neighbor-down fib-accelerate
* PFC WD
* Dtel support
* Sensor transceiver monitoring
* LLDP extended MIB: lldpremtable, lldplocporttable, lldpremmanaddrtable, lldplocmanaddrtable, lldplocporttable, lldpLocalSystemData
* Debian Kernel 4.9
* Warm Reboot
* Incremental Config (IP, LAG, Port shut/unshut)
* Asymmetric PFC
* PFC Watermark
* Routing Stack Graceful Restart
* Basic VRF and L3 VXLAN
* FRR as default routing stack
* Everflow enhancement
* Egress ACL bug fix and ACL CLI en-hancement
* L3 RIF counter support
* PMon Refactoring
* BGP-EVPN support(type 5), (related HLD Fpmsyncd,Vxlanmgr,template)