NVMe storage tailored for TCP / IP at Lightbits
With its NVMe-over-TCP solution, Lightbits Labs offers to separate storage and servers without disrupting the Ethernet network.
Founded in 2016 at Kfar Saba, the start-up Lightbits Labs intends to help companies rethink their storage architecture by making the best use of network resources in a multicloud framework. As part of a virtual IT Press Tour, due to the pandemic, Avigdor Willenz, founder and CEO of the start-up, explained to us that “the objective of providing the flexibility, elasticity and cost reduction that benefit each company using the hyperscale ”. A software-defined storage solution, Lightbits offers 100% NVMe-over-fabric flash arrays (x86 servers) that use the TCP / IP protocol for data transport. The start-up focuses above all on the performance of NVMe SSDs on the network by using NAND as a local flash memory via the NVMe / TCP protocol.
Unlike the first NVMe-oF systems to appear on the market, Lightbits’ solution does not use the RDMA over Converged Ethernet (ROCE) protocol to work. The latter, which ensures low-latency, lightweight, high-speed RDMA communication over an Ethernet network, is expensive to implement. While it provides faster data transfer for network-intensive applications, it requires a network configured for traffic without loss of information on Layer 2 alone or Layer 2 and 3. With NVMe- over TCP, existing TCP / IP cabling can be used to leverage flash arrays and reduce costs while retaining the performance and latency inherent in NVMe-oF, albeit with a few microseconds more than with ROCE.
Arrays on a standard x86 basis
“With LightOS, any data storage server equipped with NVMe SSD (QLC) can be transformed into a scalable storage solution. At the heart of LightOS is Global Flash Memory with Translation Layer (GFTL) […] with features such as logical volumes, thin provisioning, compression, and data redundancy ”explains the start-up. Access to arrays requires an operating system with a standard NVMe / TCP driver, which provides the ability to talk to arrays as if they were Direct Attached Storage (DAS). Clustering the solution ensures high availability. In the event of a failure on an SSD, online RAID protection protects the data. The services themselves are distributed for better protection. To boost their IO processing capabilities, Lightbits arrays can also be equipped with a PCIe LightField FPGA accelerator card.
Lightbits’ platform is based on standard x86 servers equipped with NVMe SSDs.
After trying to break into the SAN market with its NVMe-over-TCP solution, Lightbits is now seeking to support the growth of hyperconverged platforms. A curious position when we know that the characteristic of HCI solutions is precisely to ignore the dedicated storage arrays since the storage resources are distributed by the HCI appliances. At VMware as at Nutanix or Cisco. But the CEO’s argument is precisely to explain that today HCI solutions which lodge towards the hosting of databases like Cassandra, need more storage resources than computation.
Among the start-up’s investors are Cisco Investments, Dell Technologies Capital, Micron, SquarePeg Capital, and Walden International alongside Avigdor Willenz, Lip-Bu Tan and Marius Nacht. The start-up employs 70 people around the world, mainly in Israel and the United States.