Taipei, Taiwan – April 27, 2018 – Inventec® discloses the cutting-edge composable, disaggregated system ''Falconwitch'' and rack solution ''inRSD'', based on Intel® Rack Scale Design.
“Agility, scalability, cost effectiveness are just some of benefits of Intel® Rack Scale Design (Intel® RSD). Most importantly, Intel® RSD enables users to address emerging applications without redesigning underlying hardware”, said Jack Tsai, General Manager of Inventec Enterprise Business Group
New Era of Composable Infrastructure
Intel® RSD is an innovative way to revolutionize data center architecture. As a leading datacenter infrastructure provider, Inventec is committed to Intel® RSD solution and has designed a PCIe Gen3 based Intel® RSD v2.1 rack namely ''inRSD'' with software partner H3 Platform
, comprising of dual-socket Intel® Xeon® Scalable Processor -based K888G4 servers, RMM, Ethernet switches, and a genuine PCIe direct attached pooled IO system named ''Falconwitch''. The inRSD solution is so flexible and it can virtually support all kind of data center use cases such as NVMe storage over PCIe or fabrics, Machine Learning (ML), Network Function Virtualization (NFV), FPGA acceleration, and pooling of SSD/HDD, etc.
Fig 1: inRSD - Inventec RSD rack solution with 19” standard rack
For users who are looking for resource pooling solutions or have the taste for datacenter fabric revolution, Falconwitch can be the best fit IT infrastructure to start your own PoC.
Designed as the genuine PCIe direct attached pooled IO system based on Intel® RSD, Falconwitch can move a PCIe resource from server to server in a rack quickly, seamlessly and easily via PCIe bus with management interfaces built on open RESTful API standard.
Let’s imagine a PCIe accelerator scenario,
Accelerators such as FPGAs, GPGPUs or specialized ASICs can be very expensive. To optimize the utilization of accelerators is critical, but the workloads often vary from applications to applications and result in management difficulty. Assuming that Application A consumes 2 accelerators, and Application B consumes 4 accelerators. Traditionally, we would setup two kinds of servers - one with 2 accelerators, and the other with 4 accelerators, so that if workload becomes heavier on Application A, we could switch it to the 4-accelerator-server (see Fig 2). However, the solution overall lacks flexibility and scaling. What if we could make accelerators on-demand and move accelerators, computing, and networks close to data to reduce the latency (seen Fig 3)?
Fig 2: Traditional way to manage accelerators
Fig 3: A resource pool that help relocating accelerators with flexibility and scaling
Falconwitch with Intel RSD software is able to create ratio-optimized combination of subsystems tuned for the application. Without efforts of pulling out servers, manually install or uninstall, and re-cabling, dynamic relocations of computing resources are allowed with just a few mouse clicks, or even completed with the automation script from your own business intelligence! It offers a brilliant composable infrastructure to evolve your data center.
To further enhance its portfolio, Inventec is designing PCIe Gen4 solutions with H3 Platform. With the core design and manufacturing capabilities, the alliance can and will work closely with customers to provide cost effective, agile, high performance, and scalable Intel® RSD-based solutions.
For introduction of inRSD, the Intel® RSD based rack solution, and its fabrics, please check RSD solution
For details of the PCIe direct attached pooled IO system within inRSD, please visit Inventec Falconwitch and H3 Platform
For computing server applied within inRSD, please refer to Inventec K888G4