Fujitsu's server jujitsu: FPGA gives networking capacity fourfold boost

Chip-accelerated WAN could do wonders in the cloud

Fujitsu Labs has fitted a parallelising FPGA to a server and quadrupled effective 10Gbit/s networking capacity.

Using hardware offload to boost effective network transmission bandwidth is not new but Fujitsu's parallelising FPGA takes it to new heights.

It has developed the FPGA to speed data compression and deduplication. The FPGA has 32 parallel compression computational units, 48 parallel feature value calculation units and 32 data partitioning units. The technology delivers fresh data to these units at the appropriate times based on predictions of the completion of each calculation.

Fujitsu WAN-accelerating FPGA

Fujitsu has also reduced the amount of data sent to the FPGA. A chunk of data would first be read by the FPGA in a preprocessing phase to decide if it was compressible. If the FPGA signalled it was compressible – contained duplicate data – then the host server CPU would send the data a second time to the FPGA to be compressed.

In its latest tech, the data chunk is sent once to the FPGA, which carries out both the preprocessing and compression on it, eliminating a data reload process and saving time.

Data reload elimination

Internal Fujitsu testing using Fujitsu servers fitted with the FPGA, and the transmission of backup data and documents across a 10Gbit/s Ethernet WAN, showed effective speeds if 40Gbit/s, which it claims is an industry record.

It could be deployed in virtual appliances that can be used in cloud environments, meaning higher speed data transfers between on-premises systems, and to public clouds for data sharing, IoT log data, office data and backups.

Productisation should deliver the goods in Fujitsu's fiscal 2018. ®