I have a couple of Dell branded E2760 units (Dell MD3860f) that have only 1.6TB SSDs in them. I have noticed that they are showing the sort of deterioration in performance that I would expect to see in a consumer class desktop SSD that was running full for a while. The LUNs are showing *average* I/O latencies of 5-10ms with mixed read/write activities. Write cache hits are a rock-steady 100%, but read cache is rarely above 15-20%.
Both units are showing healthy, and their event logs are completely boring...most days no more tham maybe a battery check, or possibly a LUN verify.
I suspect that they have gotten to the state of having not enough free blocks to maintain performance, and I am looking for procedural advice on 'refreshing' them.
They are organized as 6+1R5 sets with an available net capacity of 8,939.484 GB, from which a single LUN of 8,704 GB has been provisioned. In retrospect, I am thinking that a smaller LUN should have been created there; probably no more than 8TBs, but I'm sure that someone suggested that having paid for SSDs we should make sure the biggest bang for our buck.
In one scenario I would move the contents of one of the LUNs, then destroy the LUN and possibly RAID group, and then rebuild with a larger area reserved for overprovisioning. But I want to make certain that if I go to that trouble, that I really get the disks 'clean and clear'.
Will deleting the LUN be sufficient to let the MLC SSD free all of its blocks? If not, then will deleting and reforming the RAID group do the trick? If not, then what will?
Does ~10% unallocated space sound like it will better maintain performance? I tend to use about that when using SSDs, but my lack of issues in doing that is an extremely small sample.