John Baldwin wrote: > On Thursday 18 January 2007 08:55, Randall Stewart wrote: >> Hi all: >> >> Query (with flame suit in place :-D) >> >> Currently the UMA zone's will hold all memory >> in them until .. well until the page deamon >> runs.. or so the "zone_drain()" comment says.. but >> I can't find that connection either. So I guess >> not at all :-0 > > uma_reclaim() drains all zones when it is called. It is called by the > pagedaemon when it is woken up to free some memory. > >> Should we think about adding some sort of garbage >> collector thread.. that could hang around slowly and >> periodically look for a zone with large numbers of free >> pages... and then drain that zone? > > pagedaemon is a sort of GC thread, but it kicks in whenever the system > is low on memory and asks other subsystems like UMA to free up some > memory. > Hmm.. Ok I see that path now that its been pointed out... but that means the crash I saw earlier was one where for some reason the PD did not wake up in time... Hmm.. I need to see if I can't recreate that crash ( I was a month behind and sup'd to current of course)... I would like to know the circumstances when this occurs :-) R -- Randall Stewart NSSTG - Cisco Systems Inc. 803-345-0369 <or> 803-317-4952 (cell)Received on Sun Jan 21 2007 - 09:33:13 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:39:05 UTC