Skip to main content
. Author manuscript; available in PMC: 2025 Sep 19.
Published in final edited form as: Geosci Model Dev. 2024 Sep 19;17(18):7001–7027. doi: 10.5194/gmd-17-7001-2024

Table 3.

Overview of storage options for the two HPC systems that were used for benchmarking.

Storage options
Cloud service provider Microsoft Azure Amazon Web Services
Service name CycleCloud ParallelCluster
Storage option 1 (/local) Local NVMe SSDs in RAID 0 (2 · 960 GB NVMe – block) NA
Storage option 2 (/shared) Built-in NFS: P30 tier; provisioned IOPS: 5000; provisioned throughput: 200 MB s−1 TiB−1 Elastic Block Storage (EBS) – general purpose volumes (gp3); provisioned IOPS: 3000; provisioned throughput: 1000 MiBs−1
Storage option 3 (/data) NFS file share: max IOPS 4024; burst IOPS: 10 000; throughput rate: 203 MB s−1 NA
Storage option 4 (/lustre) Lustre 150 – size 128 TB: performance profile – 150 MB s−1 TiB−1;
Lustre 250 – size 128 TB: performance profile – 250 MB s−1 TiB−1 		Lustre SCRATCH_2 option: size – 1 TB;
network throughput: 200 (1300 burst) MBs−1 TiB−1;
240 MBs−1;
disk throughput: 200 MB s−1 TiB−1 (read), 100MB s−1 TiB−1 (write)

NA: not available