I've tested Buffer Pool Extensions on an Amazon EC2 virtual machine. The VM has a single non-persistent high-performance SSD. On VM shutdown the SSD contents are lost, so it can't be used to store business data.
In order to leverage the SSD, both tempdb and buffer pool extensions were configured to use it, not unlike what was done on Azure (save the Task Scheduler-Powershell hoop jumping). Some 100 GB BPE was allocated, tempdb data files and logs are currently around 20 GB.
The system has been running nicely for about a week. Yesterday, for no apparent reason, buffer pool extensions were disabled. None of the admins admit changing the configuration.
It looks like the default trace doesn't record BPE changes, ditto for the system_health Extended Events. Unless I missed something, would there be a way to see what triggered the change?
All the Application log and Sql Server's log contain is an entry that states the pool was disabled:
Buffer pool extension has been disabled successfully. Deleting buffer pool extension "x:\sqli001.bpe".
Are there any other probable causes for disabling the BPE than human error?
There is a KB article about I/O errors that might cause loss of BPE, but the current patchlevel should include the fix, and there are no log entries about I/O errors anyway. The tempdb files have been autogrowing around the same time, but there was plenty of free space on the SSD.
The server is up-to-date with OS (Windows Server 2012 R2) and Sql Server patches (SQL Server 2014 SP1-CU6 X64 Enterprise Edition).