Paul Zarnowski [vkm AT CORNELLC.CIT.CORNELL DOT EDU] wrote:
>Mike; You seem to have a different opinion than Scott. From
>what Scott
>says, pgin/pgout activity is to be expected with ADSM. It's the
>pgsin/pgsout that I should be worried about. If you still think we are
>memory constrained, I would like to understand why (i.e., what
>numbers are you looking at?).
Paul,
I think I need to clarify a bit: Mike could very well be right that you
are memory constrained - it would really require more analysis to know for
sure, but just because you are seeing paging doesn't mean your system is
being hosed... If you have clients that are doing operations while you are
watching monitor, you are pretty much guaranteed to be doing some
persistent paging, unless your system has enough memory to cache
everything all the clients ever need (unlikely) i.e. enough physical RAM
to cache your entire DB, LOG, and any disk storage pools.
To my thinking, a system running ADSM is more like an application/database
server running something like Oracle or DB2 or whatever. It will haver
entirely different running characterists than, say, a development system.
We develop applications that run on large Oracle databases, and that's the
sort of thing I would expect to see ADSM do, especially if running on JFS
(which is how most people seem to be running Oracle that I've seen). I
could be wrong, and am open to other opinions.
Traditionally on Unix systems all paging was seen as bad. As VMM systems
have gotten more sophisticated this is no longer entirely the case. In
AIX, basically every file you load from disk via JFS will get some
persistent storage assigned to it. Loading the persistent storage from
disk requires page faults. When the persistant memory load is too high,
AIX needs to write some stuff back out to disk. More page faults are
generated. A system doing alot of JFS IO will be doing alot of
pgin/pgouts. A system that is starved for RAM will start seeing
pgsin/pgsout, altough even some of these can be healthy. On our large
production servers we like to see the CPU 100% busy and some paging. This
means the system is working more efficiently for the sort of apps we do.
For our development systems, or engineering applications this is the
opposite of what you want.
The ADSM database will require lots of random (not sequential) IO. This
means it's going to be loading alot of different blocks. Again, assuming
there is no small hotspot in the ADSM database, you will see some paging.
Now, none of this means that you don't need more RAM...
Dunno if this really helps, but hopefully it will??
-Scott
________________________________________________________________________
________________________________________________________________________
L. Scott Emmons | CableData R&D Center - El Dorado Hills, CA, USA
Staff Software Engineer| Special Projects, Systems Development Dept
(916) 939-6088 | Views and content are my own, not CableData's
|
