Storage for Cognitive Computing
The startup Vexata positions its scalable NVMe flash arrays as storage for the cognitive era. (Image: Vexata)
One year ago we introduced the Californian startup Vexata. By redesigning the storage controller architecture, the company achieves tremendous I / O throughput rates on its flash drives. Today we want to report on the development last year, in which some technical innovations were presented and ended with a reseller contract for North America with Fujitsu economically successful.
As a reminder, the structure of Vexata’s Active Data Architecture (ADA) is essentially based on the separation of data and control paths in the memory stack. “We introduced the network architecture in the data I / O,” says Ashish Gupta, Vexata’s Chief Marketing Officer, describing Vexata’s strategic defferentiation. This allows more data to be used in real time for the users because the computer CPU only has to work with the metadata. The read and write operations to and from the NVMe flash storage are handled by the Vexata router with the VX OS operating system.
In mid-2018, Vexata announced VX-OS’s new version 3.5, which claims to have achieved the performance of its in-house VX-100 flash memory “five to ten times more traditional SSDs .” His own products are thus suitable for the new age of cognitive computing.
Bottleneck Killer and Consolidation King
In August 2018, the San Jose company introduced its VMware-optimized solution. It is expected to address the evolution of modern applications that expect more bandwidth, IOPs and fast response times. “This easily overwhelms conventional all-flash arrays, especially the storage controller,” says Vexata. In addition, thanks to today’s multi-core CPUs, the number of virtual machines supported by VMware ESX Hypervisor can grow to a few hundred per rack.
“Even a single multi-core server can overwhelm the throughput of most flash memory,” the company claims. The Vexata product line VX-100F is intended to eliminate this bottleneck and also enables large scale consolidation. The capacities of NVMe-based arrays can be extended from terabytes to petabytes . They are certified for VMware ESXi 6.7 and support Microsoft Hyper V, Citrix XenServer, Oracle / RHEL Virtualization and KVM.
Fujitsu cooperation should give further impetus
In November 2018, Vexata signed a reseller agreement with Fujitsu for North America, hoping for easier market penetration. The agreement also provides for cooperation to develop vertical markets. In addition, one wants to develop a palette of reference architectures. For this, Vexata has gained access to Fujitsu Solution Labs in Sunnyvale, California. Together, they have developed a “Velocity AI” reference architecture to help users move from business intelligence to artificial intelligence with machine and deep learning.
The brand new lab employs four Primergy RX2540 Fujitsu servers with four Nvidia Tesla V100 GPUs, four 100GBe ports and Nvidia’s container-based GPU Cloud Deep Learning Stack with matching machine learning framework. Two M27 SNG 100 (100 Gbps, 32-port) switches connect to the Vexata Scale-out flash memory VX100-FS NVMe-oF, which has a capacity of 430 TB and a bandwidth of 40 GB / s.
Fujitsu expects to work together to advance the development of the new generation of IT infrastructure for its high-performance clientele. For Vexata, this collaboration means not only the recognition of technical performance, but of course, the broadening of its own customer base, including more sales. The company is still working from the $ 54 million it received as a startup, but gaining revenue every quarter. In the meantime, however, Vexata has already produced well-known clients such as Broadcom or Tata.