Comparing the storage pools in Proxmox: Start, Power, Plus and Micron

[rychy]

Прочети на български

We re-ran the fio tests on the Proxmox host itself with explicit absolute paths to each storage pool. The reason is simple: the older benchmark results were run without an absolute --filename parameter, which could point different tests at the same target and make the comparison misleading.

Correct methodology

• fio 3.39
• 4 parallel jobs
• 60 seconds
• profile: randrw with 70% read
• parameters: bs=4k, iodepth=32, direct=1
• each test is pointed at a specific pool with an absolute path

Results per pool

• vps-start: 3825 read IOPS, 1642 write IOPS, average latency about 23.5 ms
• vps-power: 12.0k read IOPS, 5157 write IOPS, average latency about 7.45 ms
• vps-plus: 6810 read IOPS, 2926 write IOPS, average latency about 13.1 ms
• micron-mirror: 4614 read IOPS, 1981 write IOPS, average latency about 19.3 ms

What this shows

• the pools do not give the same result
• the differences between them are real and measurable
• the older, similar numbers were not proof of identical plans, but of a wrongly targeted test
• once the benchmark is pointed correctly, the picture is clear

---

These results are more useful than single synthetic numbers without context, because they show how the different storage pools behave under the same methodology.

If you are looking for a VPS for real applications and want to see the other technical topics too, have a look at the VPSStorages VPS plans.

For more background on the technology see what ZFS storage is and why it matters.

[rychy]

Important clarification: the earlier, similar results should not be read as a benchmark of the individual plans. A check on Proxmox showed that the old fio commands had been run without an absolute path to the test file, which produced a misleading comparison.

Where the problem was

• fio was run with --filename=testfile without an absolute path
• this means the different tests were not guaranteed to run on the corresponding pool
• that is why the old numbers looked too similar

The correct conclusion

• not all VPS plans use the same storage profile
• a comparison between plans only makes sense if the test is pointed at the exact storage target
• the new per-pool results in the main post of this topic are more reliable

We will publish more tests with the same approach, so that there is a clear and honest picture of how the infrastructure behaves under load.