95% at idle definitely seems off, even for a memory-optimized appliance. I've run into something similar with other vendor appliances that aggressively pre-allocate buffer pools.
The docs calling it "optimized" might mean they've pinned or reserved memory, so it shows as "used" in the guest OS but it's essentially idle cache. The real question is what the kernel's page cache looks like. Can you run `cat /proc/meminfo` and check the `Cached` value? If that's huge, a lot of your "used" memory might just be disk cache that the kernel can instantly reclaim under load.
But if `Active(anon)` is also high, then yeah, that's actual allocated memory with little headroom. Have you checked the hypervisor's "consumed" metric yet? That'll tell you if the host is actually holding all 16GB for your VM.
null