Daniel Gryniewicz

There's not much you can do in Ganesha to improve I/O throughput, since Ganesha just passes I/O through to the back end. It doesn't do any kind of caching or...

@jtlayton Any ideas?

1024 for an FD seems suspicious, especially at startup. Which version of Ganesha is this?

Can you turn on FULL_DEBUG for NTIRPC? It's done like this: LOG { Components { TIRPC = FULL_DEBUG; } }

All this timeout does is determine how long we have to wait with no work before we will reduce the number of threads in the pool. I don't think 31...

Expanding a bit: Ganesha is designed to operate as an MDS, a DS, or both, for pNFS. When the feature was originally implemented, Ganesha was embedded into the data server...

pNFS never worked in Ceph. Ceph uses CRUSH to determine OSD/DS location, and CRUSH is not compatible with any published pNFS protocol. What we did before was make a variation...

No, what you want to do should be easily doable in the FlexFiles layout defined in RFC 8435.

For pNFS to work, the NFS client needs network access to the Data Servers (DSs) in addition to the MetaData Server (MDS). In normal NFS, a single Ganesha would act...

I'm not aware of any open-source implementation of a pNFS server that is actively used. In Ganesha, FSAL_GLUSTER has pNFS support in Ganesha, but I don't know anyone who has...