Would this fio test observation impede real world I/O performance?

cybertech

https://i.imgur.com/GBMt5EI.mp4

not about how much IOPS, but that during the test it drops to 0 for a while.

or its really no cause for concern?

ionswitch_stan

Open a support case - if this was my machine I wouldn't be in love with that performance.

VPS are shared resources, and that looks like some SSD's in RAID1, or on a SATA2 controller peaking at ~14k IOPS. As folks are using that system there will be some variability in the IO to your system. When there is resource contention its possible "real world" performance would be impacted.

That being said -- if you dont have a workload that is leveraging that system, it should generally be fine. If you have a latency sensitive application, or need more stable performance you likely need to pay a bit more.

cybertech

@ionswitch_stan said:
Open a support case - if this was my machine I wouldn't be in love with that performance.

VPS are shared resources, and that looks like some SSD's in RAID1, or on a SATA2 controller peaking at ~14k IOPS. As folks are using that system there will be some variability in the IO to your system. When there is resource contention its possible "real world" performance would be impacted.

That being said -- if you dont have a workload that is leveraging that system, it should generally be fine. If you have a latency sensitive application, or need more stable performance you likely need to pay a bit more.

Thanks for the explanation. this vps is set to expire; i'm just taking it as a criteria to sieve out those idlers

it did affect my use case on production (light load 0.01 really but database write sometimes take too long).

my view is that some balancing from provider is necessary as well; maybe for eg limit IOPS per guest VM if the hardware isnt gonna change/upgrade

InceptionHosting

I have had some VPS get some really ridiculous fio results on brand new NVMe sammy pro based nodes under no load and they do worse than a 2GB fujistu drive from 1996.

Under standard conditions they are fine, but an fio test runs at a snails pet snails pace.

The only thing that helped was an upgrade of the kernel version on the host node, seen this on 4 different nodes.

May be unrelated but thought I would mention it.