Dave Marquardt wrote: > "Roland" == Roland Mainz <[EMAIL PROTECTED]> writes: > > Roland> Did anyone thought about creating a tuneable in /etc/system to set the > Roland> (preferred) default page size for stack&head to something else than 8k > Roland> (e.g. 64k) globally (instead of using libmpss.so - which does not work > Roland> for inital processes, deamons started from SMF/inetd/etc., from a > empty > Roland> ENV, tmpfs and many other things...) yet ? > > Roland> Comments/ideas/suggestions welcome... > > Go poke around for auto_lpg_* variables in the kernel. The policies > can be tweaked. > > One of these days I need to write up a blog entry on how it all > works....
Any URL yet ? :-) > We've disabled LPOOB (auto large page) on x86 until we get more data > that shows it helps rather than hinders. Unfortunately 64k pages do not get used very often even when they help. For example the stack almost always uses 8k pages at the beginning even if MPSS was set to use 64k pages for the stack. The first set of pages (8*8k) is always 8k since the initial stack size isn't rounded up to a minimum size which matches the selected page size... ;-( Other examples include heap where most small allocations end-up in 8k pages (instead of the 64k ones), "tmpfs" (I could try to hack a small patch for that) and many other things. There isn't even a way to adjust the value returned by |getpagesize()|/|sysconf(_SC_PAGESIZE)| - this is all hardcoded to 8k right now (on UltraSPARC) ... ;-( > But it runs on almost all > SPARC systems (exception is UltraSPARC-III, due it's single > programmable TLB, US-III+ and beyond are fine). Yeah, I know that... I am still wondering why the designers of the original UltraSPARC-3 implemented it that way... ---- Bye, Roland -- __ . . __ (o.\ \/ /.o) [EMAIL PROTECTED] \__\/\/__/ MPEG specialist, C&&JAVA&&Sun&&Unix programmer /O /==\ O\ TEL +49 641 7950090 (;O/ \/ \O;) _______________________________________________ opensolaris-discuss mailing list opensolaris-discuss@opensolaris.org
