On 32-bit systems, generally we did not sound alarm bells if PLE (Page Life Expectancy) was above 300 seconds (5 minutes). As I have been working more and more with 64-bit systems, I catch myself wondering if there is a different "benchmark" for PLE on 64-bit systems. Recently, a colleague and I had a discussion. Although we did not come to any conclusions, I thought that I would bring forward some of the discussion points for your enjoyment.
Addresses: 64-bit systems take half as many addresses to reference the same "real estate" in memory. So this means it flsuhses twice as many "pages" as compared to 32-bit systems for the same number of addresses. Does this mean that the if PLE is above 150 on a 64-bit system that we do not need to sound the alarm?
Percentage of VAS: 32-bit systems have a 4GB Virtual Address Space (VAS). 64-bit systems do not have this limitation and may natively address much more memory. If a 32-bit system has a PLE value of 300 with 4GB, that would be less pressure in my view than the same 300 PLE value in a 64-bit system with 64 GB. This is because when a page gets in memory, I can expect it to be in there for 5 minutes. BUT many more pages are being "churned" with a PLE of 300 in the 64-bit system than the 32-bit system. Since the 64-bit system has 16x more memory (in the case above), do I sound the alarm when PLE goes below 4800?
Other factors - I am sure there are many other factors to consider in figuring this out. In my brief search of the web, I found little on this idea that the PLE value may have a different threshold than 300 in the new 64-bit world.
The SAME: It could still be the same generally accepted value of 300, but you have to wonder.
My thought is that the truth is based on sevral of these factors. I just have not had the time to fully test them out, but I hope this has been interesting. The 300 PLE value that we have known for a long time is probably based on a 4GB VAS number. The 64-bit world is not limited to that, and the PLE threshold may vary based upon amount of memory or even if AWE is turned on. AWE on 64-bit? yes - there is a case for that too althogh we won't discuss it here and now.
If someone reading this on sqlblog.com has additional information, I'd love to have it posted as a comment for discussion.
P.S. Be sure to check out www.RickSQL.com for quick "10 Minute Tidbits" about SQL Server topics.