Hello Everyone I've been evaluating glusterfs as a storage solution for ESX over the last couple of months. I have 2 debian boxes interconnected over 10gbe wmth gluster setup in a replication/mirror raid1 arrangement. glusterfs performance is excellent. Simple 'dd' tests reveal I can write at 300mb+/s to the mount. No complaints here. So I proceeded to 're export' the glusterfs mount via NFS - for service to my ESX farm. Initially I used the debian distro's nfs-kernel-server package to achieve this. NFS is reliable, however, sadly, maximum performance to any gigabit client is less than 20mb/s. If I use NFS to export a directory outside of the glusterfs mount (i.e no 're exporting' going on) I can begin to saturate the link with 90mb+/s transfers. So I switch my approach, and move to the unfs3booster userspace daemon as outlined on gluster.org. Once compiled and installed I can saturate the link on the re export of glusterfs. Fantastic! Problem solved I thought. However - a weird bug in ESX4 becomes evident . I can mount the NFS export on the ESX without a problem. It reports size and free space correctly. But I can't browse any directory that contains a large file. How large? I'm not sure exactly. I can explore directories that contain multiple small files, 1GB files are ok. But I can not list the content of directories that contain large (16GB) images/vmdk's. The datastore browser experiences a response time out when I try. There are no specific errors in the client or server syslog's that elude to why this occurs. I guess I am looking for anyone who has any experience with this type of NFS exporting to ESX. Has anyone had success with the userspace or the kernel space daemons. Do you experience the performance issues or the directory listing problem as I detail above? Incidentally, today I eliminated FUSE from the equation by switching to the booster client and the results are the same as above. If not slightly slower. Any suggestions? I am at a loss here! Thanks Matthew
