I'm confused why NFS provisioner outperforms Jiva

[madalinignisca]

Hello,

I have this scenario, deployed on dedicated servers, nvme disks in raid 1.

Installed only Hostpath, Jiva and Nfs.
Jiva uses 3 Hostpath replicas, Nfs uses Jiva.

Using a Jiva PV, IOPS are very slow.
Using Nfs, IOPS are very high, ~40% of direct NVMe.

Here are screenshots:

Any idea why is direct Jiva usage so slow?

I did benchmarks with Longhorn, which gives identical results on normal or nfs mode, performance being somewhere at the middle between Jiva and Nfs over Jiva.

[avishnu]

Thanks for being a user of OpenEBS !!
I'd like to suggest Mayastor as the alternative to Jiva.
Mayastor is a replicated storage solution (just like Jiva) but it's very fast as compared to Jiva.
Mayastor is under active development and support, unlike Jiva.

https://github.com/openebs/mayastor
https://mayastor.gitbook.io/introduction/

[madalinignisca]

I did tested occasionally Mayastor, but the CPU usage is not acceptable. The cost for CPU is highest in cloud and dedicated computing.