Michael, I responded to part of your suggestions in another post, want to complete it here.

The vendor won't tell us a lot about their hyper-convergence solution; it's patented, and I don't think their support people actually know the under-the-hood details, other than it's a black box with a really smart processor for deduping.

Their focus is on write efficiency and that, they have achieved. Watching PERFMON, using a 100 second interval, I rarely see the "Average Disk Seconds/Write" metric exceed 3 msecs, even under the highest I/O loads, using primarily 7200 rpm disk drives with a small amount of SSD and some very high speed RAM for caching, thrown in.

Reads are the only problem, and primarily on 3 files (2 application database data files, and the TempDB log file).

They claim their system is configured correctly. After personally reviewing the VMWare settings, with our admin staff and their techs involved, I'm convinced those settings are optimized.

There is something else going on, something unique to their particular implementation, that is the problem.

When our company grows, and outgrows the current system, we are almost certainly going to look for a different vendor with a different solution. Right now, we're not there. So I'm doing what my management has asked me to do: squeeze every ounce of performance out of the current system until we need to replace it.

--------

Thanks to all of you for taking the time and energy to respond.