mount | grep nfs
# or
findmnt | grep <volume-name>
Example output:
192.168.1.100:/exports/data on /var/lib/kubelet/pods/.../volumes/kubernetes.io~csi/pvc-xxx/mount
Before we dive into the "top" list, let's clarify the terminology. "Kshared" is not a universal standard but often refers to shared folders in KVM environments (using tools like virtio-fs or 9p). In other contexts, users searching for "kshared folder top" might be looking for rankings of sharing software, or they may have misspelled "Cshared" or "VBoxShared."
For this guide, we define a kshared folder as a directory on a host Linux machine (usually using a KVM hypervisor) that is made accessible to one or more guest virtual machines with high performance. kshared folder top
kubectl exec -it <pod> -- dd if=/dev/zero of=/shared-folder/test bs=1M count=100
If slow, check:
The "Folder" terminology implies organization. In a kshared implementation, the kernel exposes a hierarchical namespace. mount | grep nfs # or findmnt | grep <volume-name>
You need a combination:
| Tool | Purpose |
|------|---------|
| iostat | Monitor NFS/client I/O on node |
| nfsiostat | NFS-specific stats (latency, ops) |
| iotop | Per-process disk I/O (limited with network fs) |
| fio | Benchmark shared folder |
| kubectl exec -- df -h | Check mount usage |
| node_exporter + Prometheus | Collect NFS/client metrics |
| lsof | See which pods have files open in shared folder | Example output:
192
Ideal command (if it existed):
kshared-top --namespace=default --volume=pvc-xyz → shows pod → read/write MB/s, ops/sec, latency.
We’ll simulate that.
Best for: Legacy kernels or simple read-only shares. Top limitation: Poor performance with many small files. Avoid for Git repositories.
As of late 2024, the "top" development on the horizon is virtio-fs with vDPA (vhost Data Path Acceleration) . This will allow the kshared folder to bypass the VMM entirely, offering bare-metal I/O speeds for shared memory regions. Additionally, NVIDIA is working on GPU-Direct storage passthrough for shared folders, meaning your VM could train AI models directly on host-stored datasets without copying them first.