On Fri, Sep 9, 2016 at 10:30 PM, Fernando Frediani
<fernando.frediani(a)upx.com.br> wrote:
Hi there.
I was reading this interesting URL someone just sent a while ago regarding
hyperconvergence topic
(
https://www.ovirt.org/blog/2016/08/up-and-running-with-ovirt-4-0-and-glus...)
and found the point about the optimal amount of resources for a Engine -
16GB of RAM.
I just wanted to ask what component or feature eats up so much memory for
that amount be the recommended. Or is it just in a hyperconverged scenario ?
The engine itself needs quite a lot of memory.
Are there any components that can be optional that can reduce the amount of
memory needed to run the Engine ?
DWH and postgresql need quite a lot too.
In principle you can run each of: engine, dwh, engine's db, dwh's db
on its own machine, total of 4 machines.
Also please note that the "minimum" according to engine-setup (and
elsewhere) is 4GB, and that this greatly depends on your setup's size
and use patterns/flow (e.g. how often do you ask the engine to do
something compared to the engine most of the time just monitors stuff
etc).
Also with if the Data Warehouse runs in a separate host what would be the
reduction in resources consumption, specially memory ?
I am not aware of actual semi-accurate measurements done recently,
and I guess most people don't bother because memory is cheap these
days.
Personally I usually run my engines on VMs with 2-4GB RAM, and
never had problems. These are used only for testing/development
and not for real work. I am also aware of engines on machines
with 8GB RAM, managing 100-200 VMs with no problems.
See also:
https://bugzilla.redhat.com/show_bug.cgi?id=1185411
https://bugzilla.redhat.com/show_bug.cgi?id=1329119
Bottom line:
If you don't care much and have a few tens/low hundreds VMs,
just use 8-16GB.
If you do care, and/or have much larger setups, either separate
the services to different machines, or actually do some tests
(and publish the results!), or both.
Hope this helps,
--
Didi