IoT sensors are generating a lot of data. In large scale settings, data collection by gateways need AI based aggregators. AG1000 fulfils the need for data aggregation at the edge. We created this scalable array server containing 4 NVIDIA® Jetson™ TX2 modules. The server array connect via Gigabit Ethernet fabric through 10G Uplink interface. The aggregator is a modular system supports 4 front panel cards and a back panel computing and storage engine.
The system uses high speed PCIe express switch and connects different modules with it. It has on board ARM based computing module with 10 Gbps uplink and management interface. It also has a serial console port for on board BMC system access and configuration. OpenBMC designed by Facebook’s openCompute project has been implemented which has 256 MB of RAM and 32 MB of Flash Storage. It is a secured system on a chip.
The system provides upto 8 Teraflops processing. This enable super computer like features at the edge to perform machine learning by applying TensorFlow. Local computing and storage is used to process data before sending to central servers for big data analytics.
There are four slots in the front panel accommodating four different modules. The four modules are – 24 Port gigabit ethernet, 8 port fiber optic, 8 Port HDMI and high capacity software defined networking module.
At the back panel, an extra computing module can be added with two SSD disk drives.
Gigabit Ethernet Module: A 24 Port Gigabit Ethernet / 1000BASE-T / IEEE 802.3ab switching module which can provide at least 24 other gateways to connect to each module. Using all four front panel modules as Ethernet ports, it can aggregate 96 Gigabit Ethernet ports. Our AG1000 can combine at least 960 iG100 gateways or 96 iG200 gateways.
Fiber Module: This contain 8 port fiber optic ports in one module. 6 of the ports are SFP 1G and 2 ports are SFP + 10G ports. If all four ports are used as fiber, total 32 ports fiber can be added. Currently significant work has been in underway to add 40 Gb and 100 Gb interface. Mellanox EDR module is also under development
Software Defined Network: The SDR module supports 100kHz – 3.8GHz frequency range which supports 120 MHz RF bandwidth. This RF module can scan the network and apply machine learning for monitoring and detecting signals, decoding information and creating alarms if it finds any anomalies. FPGA based SDR provides all three RATs (Radio Access Technologies) – 2G, 3G, and LTE frequency scanning. It can be used to detect fake BTSs for mitigation purpose. The system can be used for situational aware radar system and many other RF signal related use cases.
Video Input Module: There has been high demand of video analytics which includes live broadcasting monitoring. Our Video input module supports all live television feeds, provide voice synthesis including language translations. We also offer live sentiment analysis based on natural language processing using this module.
Computing Module: An additional computing module can be connected at the back side of the AG1000 which is supports COM express type 6 module inside the module. Currently, the system supports 2 SATA ports as well as PCIe express slot for SSD drives, VGA port, COM port, console port, 2 Ethernet LAN and one management LAN with 2 USB port for connecting keyboard and mouse. The system also supports 10G Ethernet port which connects to the main board. Additional management Ethernet ports are also provided in the additional computing module
Computing and Processing at the Edge
AG1000 is self-contained and do all computing and processing itself, no external units required.
AI and Machine Learning at the Edge
AG1000 have Artificial Intelligence and Machine Learning built-in to this device. No need to attach external device or send data somewhere else.
Highly Expandable and Scalable
You can daisy chain multiple AG1000 to expand it’s capability and capacity.