I don't necessarily disagree with any of this discussion (auto memory
mgmt can certainly thrash sometimes) however I'd offer two additions:
1) I'm currently using ASMM across a rather large estate. Since I
arrived a few years ago, we have been following a policy where we only
set the minimums if there are problems. So far we haven't had to set
a minimum yet. We do have some very busy databases, although
admittedly we also write the app that runs on them so things mostly
use binds and such, avoiding some of the more common exploding shared
pool issues. Also our devs work great with our DBAs - when we see
potential problems they are usually right on top of making
improvements in the app to address them.
2) versions please, as Oracle is often changing things between
releases. I have a lot of systems running on 11.2.0.3 with ASMM and
we haven't yet seen thrashing (previous caveat may contribute).
That said, I don't think minimums are a bad idea. Although I'm not
sure if it would really prevent "thrashing" (seems like that's usually
jumping back and forth over some small threshold) but rather just
prevents outright stupid allocations like Ls Cheng mentioned...
-J
--
http://about.me/jeremy_schneider
On Wed, Apr 15, 2015 at 3:03 AM, Ls Cheng <exriscer@xxxxxxxxx> wrote:
Hi--
I dont use AMM at all but do use ASMM. With ASMM I set minimum values for
all pool and cache, usually leave 10% to 20% free so the database kernel can
manage
On Wed, Apr 15, 2015 at 12:30 AM, MARK BRINSMEAD <mark.brinsmead@xxxxxxxxx>
wrote:
That sort of thing is hardly shocking, if you set AMM loose and let it do
its own thing completely without constraints. Sadly, though, that seems to
be more-or-less the way Oracle had envisioned its use.
You'll generally be much happier with AMM if you set lower-bounds for all
of your major memory pools, and leave AMM less room to play around. In my
mind, leaving (only) about 20% of the database memory to AMMs discretion
seems about right. I have no rigorous math or case-studies to support that
-- just instinct.
Perhaps others have studied this more thoroughly?
On Tue, Apr 14, 2015 at 5:59 PM, Ls Cheng <exriscer@xxxxxxxxx> wrote:
+1
AMM is a piece of junk, last customer had 8GB memory_target and after a
week AMM set db_cache_size to 64M.. uh oh
