Hi Benjamin,
If your primary concern is performance, over availability, any of the options would suffice, assuming that the SSDs are of a similar type and the capacity is the only major difference. Each of the disk groups you would be building with your options have the same HDD/SSD capacity ratio of ~12:1, which for most workloads would be adequate. You performance will be primarily dictated by the speed and latency of your SSD devices. The HDD's performance will only really come into play when you are experiencing read cache misses. The 4th setup would give you the best read performance in the case of cache misses, therefore, as you have specified 3 HDDs and one SSD, resulting in a single diskgroup with 3 spindles.
However, I would argue that another configuration would be even better, if the costs, and the server hardware allows for it. It looks like you want 12TB of capacity and 960GB SSD per node. Could you go for a configuration like the following?
2 x 480GB SSD
4 x 3TB HDD (or 6 x 2TB HDD)
That means you have the same ratio of SSD:HDD capacity, but greater redundancy and more failure domains. i.e EAch host will now create 2 diskgroups, with one SSD for cache and multiple HDDs for capacity. This way, a disk failure will less impact on your cluster, while still providing roughly the same performance characteristics.
Each disk will also have their own queue depth, so more disks allows for a greater overall queue depth at the disk level, allowing for potentially better performance.
Before making any decisions though, you have to consider future growth also. If you fill all the drive bays with smaller capacity disks, you may not have the option to add disks later to scale up. You may be forced to scale out by purchasing more servers.
Lots to consider!
