Page Life Expectancy drops to zero even if buffer has pages that are several minutes old

Question

Scenario: SQL 2014 with 10 GB allocated memory. There are three tables in a database with 3 GB data in each. PLE is pretty high(1000s).
I query first table, it gets loaded into memory and takes up 3 GB, verified by sys.dm_os_buffer_descriptors. PLE keeps rising at the rate of 1 per second. I query second table, it gets loaded into memory and takes up another 3 GB. PLE still keeps going
up at the rate of 1 per second. I query the third table, it gets loaded into memory and takes up 3 GB. SQL flushes out some pages of first table from memory and the first table is now occupying only 500 MB of memory. So, SQL did not flush out all pages
of first table. The difference between time when the first table was loaded and when the third table was loaded was more than 5 minutes. However, when the third table was loaded, PLE went down to close to 0 even though there
are pages from first table in memory that are over 5 minutes old. Does someone know why PLE would go down to close to 0? I thought it would go down to somewhere around 300(5 minutes). 
Note: This was done on a test server with no one else using the server and no other app on the server. So, no interference.

Answer 1

Hi,
There was bug in SQL Server 2012 which caused this behavior. Please read this
support article reason given by MS was
The resource monitor and lazy writer activities in SQL Server are designed to maintain cache sizes and free list capacities to avoid memory shortage situations. Under rare circumstances, the memory manager in the SQL Server database engine may enable
the SHRINK state for the Buffer Pool for longer than required. When approaching the memory target, a sudden burst of memory allocations may cause memory allocation failures and internal retries. During the retry activities, the memory manager can enable
SHRINK states on the various caches in order to support the new memory requests. Under these retries there is a small window in which the SHRINK state may remain enabled longer than necessary to restore the free list capacities.
This causes a temporary release of more memory than is required to return to a steady state. For example, the buffer pool can release more data pages than required, inflating the free list depths and reducing the page life expectancy (PLE) for the buffer
pool. In this situation, you'll notice a quick decline in the SQLServer:Buffer Manager\Page life expectancy
performance counter values and a quick increase in SQLServer:Memory Manager\Free Memory (KB) performance counter values.
I am not sure you are hitting this behavior in SQL Server 2014.
Please mark this reply as answer if it solved your issue or vote as helpful if it helped so that other forum members can benefit from it
My Technet Wiki Article
MVP