This seems to be one of most favorite questions ever: "How come that Exchange Store.exe keeps on grabbing more and more memory, even on the server that is not very busy? Is there a memory leak?". T...
Updated Jul 01, 2019
Version 2.0
Blog Post
What Nino tried to explain above (memory backoff alogrithm) is called affectionately DBA which stands for Dynamic Buffer Allocation. Basically depending on the memory pressure being put on the system (and being sampled at a fairly aggressive rate), the store will release memory back to the system. For example, go to your Exchange server and start up IE or MMC. Both spike memory usage at thier load time. If you are watching the memory allocated to each process you will see Exchange store give up memory to the system (so it can then turn around and give it to IE / MMC). You will have to trust me.. this is how it works.
Now it wont release ALL of it at once, but will release it over (a relatively short) time. On the flip side, we don't aggressively eat it as it becomes available. Again based on memory pressure we will consume free memory in chunks. Think of it as Exchange's private memory manager that takes input from the entire system and is a benevolent participant.
The reason for doing it gradually is that you don't end up in a seesaw effect on the server where you are reacting to every absolute request for memory etc. Realize that the "memory" you are evicting is essentially the database cache and if you evict too aggressively and then re-allocate you end up with unpredictable and in-efficient IO patterns and your hit ratio on the cache drops (plus all the added IO's required to flush the cache pages to disk and the subsequent reads to fill the cache).
This now explains why on an Exchange server unused memory is essentially a wasted resource and if you can effectively approprirate and evict the resource, you net make your Exchange server as efficient as possible.
Now it wont release ALL of it at once, but will release it over (a relatively short) time. On the flip side, we don't aggressively eat it as it becomes available. Again based on memory pressure we will consume free memory in chunks. Think of it as Exchange's private memory manager that takes input from the entire system and is a benevolent participant.
The reason for doing it gradually is that you don't end up in a seesaw effect on the server where you are reacting to every absolute request for memory etc. Realize that the "memory" you are evicting is essentially the database cache and if you evict too aggressively and then re-allocate you end up with unpredictable and in-efficient IO patterns and your hit ratio on the cache drops (plus all the added IO's required to flush the cache pages to disk and the subsequent reads to fill the cache).
This now explains why on an Exchange server unused memory is essentially a wasted resource and if you can effectively approprirate and evict the resource, you net make your Exchange server as efficient as possible.