LJI is an independent research institution with a 30-year focus on analysis of immune system disease at the atomic, genetic, protein and cellular level – analyses that are fundamental across hundreds of diseases. By early 2020 LJI’s storage architecture had grown to5PB of raw data and nearly 850 million files. As COVID-19 boosted research by its multi-lab Coronavirus Taskforce, and use of its Krios cryo-electron microscope expanded, LJI’s data volumes were skyrocketing. Intensive sequencing workloads often involved 1 million reads per second, speeds that could swamp the limited local storage of the GPUs. With growing data sets, the Institute’s IT team needed to provide scale-out external storage with higher performance to support continued timely analyses to its over 200 Ph.D.-level scientists, informaticists, and research staff.
“We were looking for blazing storage IOPs and throughput to enable researchers to use Krios without IT constraints, and we certainly found it with the solution offered by Excelero, which used KIOXIA CD Series SSDs,” said Michael Scarpelli, senior director of information technology at LJI. “We can’t allow any limits on IT resources to bottleneck the fundamental research that is vital to finding treatments for COVID-19 and so many more diseases. With the Excelero-driven system, we never even have to worry about performance. The system just flies,” Scarpelli said.
“The storage system's critical feature is its custom database application, and Excelero has boosted transactions per second by 10X with that database alone. It’s simply phenomenal,” Scarpelli continued.
Supporting both a standard high-performance computing (HPC) cluster and the GPU cluster, LJI’s storage infrastructure replaced a just a bunch of flash (JBOF) deployment of conventional Flash drives that was difficult to scale. Designed to serve the superior processing speed of GPU servers, LJI’s new architecture was a harmonious answer to what analysts call the “GPU storage bottleneck,” where the up to 2 million IOPS capability of the GPU vastly exceeds the 400K IOPS that local storage can feed it, and can increase time to insights.
The new high-performance storage system leverages Excelero NVMesh® to enable an accelerated software-defined storage stack that connects CPU and GPU compute servers to NVMe over the network and delivers performance that is unachievable in local storage deployments alone.LJI now can virtualise NVMe storage assets and deliver them as if they were local and with limitless scale. NVMesh is the elastic block storage engine that now speeds the revamped file server environment. This ensures the many complex interactions with petabyte-scale data sets will not be bottlenecked by storage infrastructure.
Based on workload, the Excelero storage architecture enables up to a 10X improvement in price/performance. Because NVMesh is hardware-agnostic, LJI now has a future-proof storage architecture enabling the rapid introduction of modern NVMe, network and server technologies without delays often introduced by rigid appliance-based solutions.
LJI chose the KIOXIA CD Series data centre NVMe SSDs for the flash storage portion of this project. “KIOXIA’s mission is to uplift the world with memory, and our data centre SSDs are designed to deliver a balance of performance, low latency and data protection,” said Jeremy Werner, senior vice president and general manager, SSD business unit at KIOXIA America, Inc. “We are pleased that Excelero and La Jolla Institute chose to use our drives and have obtained such success.”
“The life-saving research of the La Jolla Institute should have nothing stand in its way. High performance research workloads need much more storage capacity than that available locally in GPU systems. By incorporating Excelero NVMesh into its GPU cluster and storage architecture, LJI can now achieve the highest-performance at any scale, with the best price/performance,” said Yaniv Romem, CEO of Excelero. “Excelero is proud to help LJI expedite its mission by optimizing its data storage.”