Case Study: SciNet achieves faster checkpointing and high-performance burst buffering with Excelero NVMesh

Pooling Nvme Within Gpfs Nsds Enables Efficient Burst Buffer

SciNet, Canada's largest supercomputer center, needed to reduce the time its new supercomputer spent on checkpointing to meet strict availability SLAs. Its challenge was to complete these large-scale checkpoints within a 15-minute window. SciNet turned to Excelero and its NVMesh software to implement a high-performance burst buffer solution.

Excelero's NVMesh enabled SciNet to create a petabyte-scale, unified pool of NVMe flash storage from 80 devices across just 10 servers. This solution delivered exceptional performance with 148 GB/s of write throughput and over 20 million random 4k IOPS. By implementing NVMesh, SciNet successfully met its critical 15-minute checkpoint window, achieving an extremely cost-effective and high-bandwidth burst buffer that ensured the supercomputer's availability.

SciNet