Hello Everyone. We received excellent feedback from the previous storage performance investigations, particularly the technotes on optimal disk configuration settings (i.e., aio native, io_uring, and iothreads) and the deep dive into optimizing guest storage latency.
Several community members asked us to quantify the difference between iSCSI and NVMe/TCP initiators in Proxmox. So, we ran a battery of tests using both protocols and here's the TLDR:
As always, questions, corrections, and ideas for new experiments are welcome.
Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox
Several community members asked us to quantify the difference between iSCSI and NVMe/TCP initiators in Proxmox. So, we ran a battery of tests using both protocols and here's the TLDR:
- In almost all workloads, NVMe/TCP outperforms iSCSI in terms of IOPS while simultaneously offering lower latency.
- For workloads with smaller I/O sizes, you can expect an IOPS improvement of 30% and a latency improvement of 20%.
- Workloads with little or no concurrency (i.e., QD1) see an 18% performance improvement.
- For 4K workloads, peak IOPS gains are 51% with 34% lower latency.
As always, questions, corrections, and ideas for new experiments are welcome.
Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox