ScaleFlux, new iteration for compute + storage approach

ScaleFlux (www.scaleflux.com), recent player in the "new old" storage board approach, recently participated to DataWorks Summit 2017 in San Jose where they exhibit their solution on the Inspur booth.
The company was incorporated in 2014 and has two headquarters, for US in San Jose, California, and for Asia in Beijing, China. We met some executives at that show and globally three executives are listed on the web site: Mr. Hao Zhong, CEO and Director, Mr. Jian/Jeff Wu, COO, he was before at Pactera for most of his career, and Mr. Thad Omura, VP of Business Development. He came from Seagate, LSI, SandForce, Mellanox, Freescale and Marvell and is advisor at Liqid. Globally we found 18 people on LinkedIn.
According to PitchBook, ScaleFlux has raised so far $12 million in 2 rounds from 5 investors. The company designs a software and flash-based hardware subsystem produced as a board and available in PCIe HHHL or U.2 form factor.

The storage entity is built with 3D NAND chips and the final product offers variable capacities from 1.6 to 6.4TB. These devices appear as block devices for the operating system. The table below presents more details.

Device	Form factor	Capacity (TB)	Random Sustained R/W K IO/s	Sequential R/W MB/s	Max Power	Endurance DWPD
CSSP2Px016B0	PCIe HHHL Card	1.6	500/100	3000/2000	25	3/5/10
CSSP2Px032B0	PCIe HHHL Card	3.2	600/150	3000/2400	25	3/5/10
CSSP2Px064B0	PCIe HHHL Card	6.4	600/150	3000/2800	25	3/5/10
CSSU2Px016B0	U.2	1.6	500/100	3000/2000	25	3/5/10
CSSU2Px032B0	U.2	3.2	600/150	3000/2400	25	3/5/10
CSSU2Px064B0	U.2	6.4	600/150	3000/2800	25	3/5/10

They name this computing model, the Computational Storage Subsystem aka CSS and they promote currently the first iteration with the 1000 Series.

The main idea is to offload and deport from central cores certain tasks that can be executed by dedicated peripheral engines. At the same time, these compute engines leverage the fast and direct connection to the Flash storage. Functions could be things that are data intensive related to erasure coding, compression, encryption and KV store. CSS is available for x86/Linux server and fully addressable via ScaleFlux software module. The engines are exposed via APIs offered by this SW module. Interesting idea, let's see how the market and partners will adopt it...