I've been doing some cursory research on ZFS tiering and it seems to come down to ARC and log cache on top of slower drives. It started me thinking about our current environment and what would be the best approach to our storage.
We were looking at moving our databases to SSD arrays with the remaining VMs on our spinning rust mirrored pool. Granted our Napp-It servers are actually giving decent (for us) performance on our spinners but we want the databases to have a little more throughput. I also don't like the idea of putting the databases on the SSD "just because" since they are not always going to be the performance bottleneck.
Right now our main array is 24TB using mirrored 2TB drive (DT01ACA200). The SSD array is a RAID-Z2 array of 960GB Sandisk Pros (6.6TB). There's actually a 3rd array which is just for running our backup environment (PHD and Veeam) which doesn't factor in but is included for the sake of being complete.
Based on our current hardware would it make more sense to run the SSDs as L2ARC on top of the spinners and let ZFS handle the caching or are we better off running 2 different NFS endpoints?
Our SSD array has 72GB of RAM (physical) and our hard drive array has 48GB of RAM (virtual).
This post ZFS L2ARC (Brendan Gregg) really got me thinking about this more and makes me wonder what would be the best.
Of course the lazy side of me (devil) says to use the SSDs as L2ARC and let ZFS sort it out. Even the angel side says in this case the devil has a good point.
Our databases are generally very read heavy outside of specific points in the month where we handle bulk data loads. It's mostly reporting and analysis work using SQL Server.
I'm still looking for additional SSD heavy L2ARC comparisons but hopefully some of the big data guys here can weigh in.
We were looking at moving our databases to SSD arrays with the remaining VMs on our spinning rust mirrored pool. Granted our Napp-It servers are actually giving decent (for us) performance on our spinners but we want the databases to have a little more throughput. I also don't like the idea of putting the databases on the SSD "just because" since they are not always going to be the performance bottleneck.
Right now our main array is 24TB using mirrored 2TB drive (DT01ACA200). The SSD array is a RAID-Z2 array of 960GB Sandisk Pros (6.6TB). There's actually a 3rd array which is just for running our backup environment (PHD and Veeam) which doesn't factor in but is included for the sake of being complete.
Based on our current hardware would it make more sense to run the SSDs as L2ARC on top of the spinners and let ZFS handle the caching or are we better off running 2 different NFS endpoints?
Our SSD array has 72GB of RAM (physical) and our hard drive array has 48GB of RAM (virtual).
This post ZFS L2ARC (Brendan Gregg) really got me thinking about this more and makes me wonder what would be the best.
Of course the lazy side of me (devil) says to use the SSDs as L2ARC and let ZFS sort it out. Even the angel side says in this case the devil has a good point.
Our databases are generally very read heavy outside of specific points in the month where we handle bulk data loads. It's mostly reporting and analysis work using SQL Server.
I'm still looking for additional SSD heavy L2ARC comparisons but hopefully some of the big data guys here can weigh in.
