ITPT #29: Vexata is getting stronger
by Bertrand Garé , December 14, 2018 13:11
We had met Vexata a year ago. We took advantage of this new “tour” to redo a point on the activity of this company. The chosen architecture is still a model of its kind and perfectly matches the storage loads required by high-level analytical applications. We are not the only ones to believe it, Fujitsu has placed Vexata as a reference solution in the field in its North American laboratory with reference architectures on its Primergy servers.
Vexata is known to allow very high I / O performance compared to more traditional storage solutions. Its architecture makes the difference on this point by eliminating the overly intensive use of CPUs by routing data differently and confining CPUs only to operations on metadata. This “Active Data” architecture benefits from hardware acceleration, and distributed parallelism to handle the I/O, algorithms to treat the scheduling of I/O on each local SSD and metadata processing in. These features make the architecture align perfectly with the needs of data storage for analytics applications.
A partnership with Fujitsu
These qualities were noted by Fujitsu, which integrates the Vexata solution into its new North American laboratory by offering demonstrations of this solution on reference architectures based on Fujitsu’s Primergy servers. This partnership should eventually extend to other geographical areas. Fujitsu sees the opportunity to truly change storage economics for machine or deep learning applications, which will help customers move from simple analytical applications to cognitive solutions. These solutions can be optimized in different verticals requiring specific performance.
The reference architecture proposed in the laboratory relies on 4 Primergy RX2540 servers with 4 Nvidia Tesla V100 GPUs, 4 ports with 100 Gb operating containers embedding the deep learning software stack in the Nvidia Cloud. The connectivity between the servers and Vexata is realized by 2 Mellanox SN 2700 switches with 32 ports of 100 Gbps Ethernet. For storage the solution uses the Vexata VX-100FS NVMe-oF array with a capacity of 430 TB and a bandwidth of 40 Gb/s. The extension of the solution is carried out horizontally.
During our visit we were treated to a demonstration that was based on a Vexata VX 100-F full flash bay with 16 RAID 5 storage modules connected by a 16-port Fiber Channel switch to Primergy servers RX2540 M4 with Intel Xeon Gold 6154 CPUs at 3.00GHz and 192 GB of memory attached to a Red Hat server that was based on an Oracle 12 C base on 4 nodes. The Primergy server was connected to the base with 40 Gb connectivity.