Hi All,
Is there any known issues or performance related tweaks with regards to clone performance. We are finding cloned volumes mounted as read only (for backup purposes) much slower to read from than regular volumes.
Our setup all uses very good hardware (dual Intel CPU, 256GB RAM, 24x NVME disks, 40GB NIC) - we have 8 in total.
These are used as the SANs for a Hyper-V Cluster (about 40 front end nodes).
All SANs and Front End Nodes are connected via iSCSI to our SAN Network switches at 40GB.
All iSCSI volumes are formatted with NTFS.
We have been testing two different backup methods:
1. On the Hyper-V node that owns the iSCSI volume we make a shadow copy, and then use UNC to backup files on the volume to our backupserver (also connected via 40GB).
2. In TrueNAS create a snapshot, then read-only clone, serve that directly via iSCSI to the backupserver.
For some reason beyond my understanding performance is at least 30-40% better using option 1. The traffic actually has to travel from the SAN -iscsi> Node -unc> BackupServer - how this considerably outperforms the clone directly connected via iSCSI to the backupserver (SAN -iscsi> BackupServer) I do not understand. This occurs across TrueNAS and Nexenta 5, so I believe it has to be an OpenZFS issue.
Does anyone have any experience with this, or any properties or configuration that may help?
Thanks,
Will Kruss
Is there any known issues or performance related tweaks with regards to clone performance. We are finding cloned volumes mounted as read only (for backup purposes) much slower to read from than regular volumes.
Our setup all uses very good hardware (dual Intel CPU, 256GB RAM, 24x NVME disks, 40GB NIC) - we have 8 in total.
These are used as the SANs for a Hyper-V Cluster (about 40 front end nodes).
All SANs and Front End Nodes are connected via iSCSI to our SAN Network switches at 40GB.
All iSCSI volumes are formatted with NTFS.
We have been testing two different backup methods:
1. On the Hyper-V node that owns the iSCSI volume we make a shadow copy, and then use UNC to backup files on the volume to our backupserver (also connected via 40GB).
2. In TrueNAS create a snapshot, then read-only clone, serve that directly via iSCSI to the backupserver.
For some reason beyond my understanding performance is at least 30-40% better using option 1. The traffic actually has to travel from the SAN -iscsi> Node -unc> BackupServer - how this considerably outperforms the clone directly connected via iSCSI to the backupserver (SAN -iscsi> BackupServer) I do not understand. This occurs across TrueNAS and Nexenta 5, so I believe it has to be an OpenZFS issue.
Does anyone have any experience with this, or any properties or configuration that may help?
Thanks,
Will Kruss
