Intel has partnered with Lightbits Labs (investing in the start-up alongside Dell and Cisco), to improve the performance of storage systems in data centers. The two companies plan to develop disaggregated solutions to reduce the total cost of ownership (TCO) of storage systems, including limiting unnecessary capacity allocations. By refining the allocation of storage resources, Lightbits intends to allocate just what is necessary for applications and VMs. The end result is – in theory at least – less power consumption since unused space is no longer allocated. It is always difficult – unless you have dedicated analytical tools – to optimize the forecast allocation of storage resources. The cloud can however help to refine this management of the available or reserved space in hybrid mode.
Based in San Jose, California, and founded in 2016 in Kfar Saba , Israel, the start-up sees itself as the “creator” of NVMe-over-TCP because it contributed code and worked alongside Facebook, Intel, Cisco, Dell EMC and Micron to develop the NVMe-over-TCP standard, which was ratified by the NVM Express consortium at the end of 2018 as a link transport layer. Lightbits also aims to improve TCO by eliminating the need for specialized hardware. With LightOS NVMe over Fabrics TCP (NVMe-oF / TCP), the start-up has indeed developed a software-defined block storage platform that groups together NVMe flash storage space distributed over several nodes over Ethernet using the TCP / protocol. IP rather than a more expensive specialized network such as Fiber Channel.
Faster SSDs in NVMe
Intel, which markets NVMe SSDs with NAND flash or Optane components is probably not the first name that comes to mind in the small storage world, but it does work with major vendors. The founder explains that LightOS, while being fully optimized for Intel hardware, will provide customers with improved storage efficiency, reduce underutilization, and maintain compatibility with existing infrastructure without compromising performance.
“Data centers are undergoing transformation, with disaggregation and composability of resources being essential to meet the efficiency demands needed to cope with the data explosion,” said Rémi EL-Ouazzane, vice president and director of strategy and business development for the data platforms group at Intel, in a press release. “Our differentiated hardware capabilities combined with Lightbits’ innovative NVMe over Fabrics software provide our common customers with a cost effective and exceptional solution to address this strategic inflection point.”
Intel equipment to start
The LightOS storage solution has been tested with Intel Ethernet 800 Network Adapters with Application Device Queues (ADQ), designed to achieve faster and more predictable Ethernet. ADQ enables NVMe-oF / TCP interface to achieve distributed storage performance in the same range as RDMA-based protocols, while NVMe-oF / TCP allows wide adoption due to its ease of deployment and its scalability. Intel claims that LightOS with ADQ on the Ethernet 800 card shows up to 30% improvement in predictability of response time measured by P99.99 queue latency, up to 50% reduction in average latency, and up to 50% reduction in average latency. ‘70% increase in throughput measured in IOPS when using ADQ vs. without ADQ. In addition to the Intel Ethernet Adapter,
Lightbits Labs isn’t the first NVME-oF storage startup that Intel is investing in. This was already the case for the promising young shoot E8 Storage , whom we met several times within the framework of the IT Press Tour, and which ended up in the hands of AWS. , but did not disclose the financial details of its investment. Protocols like NVMe and NVMe-oF enable ultra-fast, low-latency flash storage, which means businesses can run distributed applications, including Kubernetes, and access data in near real time.
