On Thu, Apr 2, 2009 at 8:15 AM, Shane Ginnane <[email protected]> wrote:
> This is an indication of preference only - in a stress situation your > input will be ignored. Well it should be - I have trouble explaining what > Marcy saw at the setting of 60. My opinion* is that the two memory management algorithms interfere. There's probably Java code implementing various type of data cache classes in the JVM heap, likely on LRU basis. Linux will select candidates for swap-out also on LRU basis. This typically results in worst case scenario where the page selected for swap-out is the next one to be re-used by the application. When the page is re-used, the swap splot remains assigned until another swap-out for that page happens. So when this popular part of application memory just does not fit in what Linux set aside for it, you will see it over time cycle through. Resulting in every page eventually having a slot assigned in swap space. So it fills up swap space. When she lowered swappiness enough, it shifted the base line and left enough room for the popular part of the application without swap out the least recently used pages all the time. So it did not allocate all these swap slots and did not fill up swap space. *opinion: My best guess of what happened without measurements to prove it Rob -- Rob van der Heij Velocity Software http://www.velocitysoftware.com/ ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
