I'm building a GlusterFS share for a heavy Get Attribute workload. This cluster is configured in "Replicated fashion" and needs to host several TB of data.
In this scenario the largest part of request from clients is of type "Get Attribute" ( eg. Apache last modified ).
I'm wondering about IO ( disk ) performance:
- How does GlusterFS performs in terms of IO on disks for Get Attribute?
- Does each Get Attribute request reach the underlying disk or are they served from the server without impacting disk IO?
- Is gluster caching effective in this scenario?
- Is there a request to the replica server each time a Get Attribute reaches the Gluster Server?
- Considering Gluster as the preferred server solution, is it better to use GlusterFS client of NFS client under this specific workload ?
Thanks in advance.
