I would do empirical benchmarks, but it's tricky on Amazon instances, my
experience is that the actual performance can vary a lot over the course
of a month. The problem is that these units don't do much accounting on
I/O or memory bandwidth, so it all depends on what your neighbors are doing.
What you could do is setup JMX and observe the machine with visualvm
(tool included with Java SE to monitor Java Virtual machines). If it's
constantly doing garbage collection you need to give it more heap. If it
doesn't, then keep Xmx where it is, or reduce it a bit. That would also
give you a clue whether you need more RAM, or should get more CPU instead.
Once you've given it adequate heap (remember this can change depending
on the number and types of requests), you can do the math and see if
your data on disk fits into the leftover RAM. The numbers mentioned
earlier (139 for buffer, 352 for cache) don't really tell me much, but
it doesn't seem squeezed.
-Arne
On 11/27/10 3:35 AM, Chris Holmes wrote:
How much data is going to be served up? The advantage of more memory
would be that your shapefile disk block would make it in to the OS
disk cache, and then things would go quite fast. Of course once that
happens then you'll definitely be CPU bound, so if most of your data
will fit in the 7g of ram then the more cpus will be faster. Arne
knows this stuff much better than I, and he's the one I learned from,
but giving less explicit memory to java and postgis may help (or maybe
it's smarter than that).
I guess the other thing to consider is how many concurrent users.
Even with tiling I think you don't get more than 4 threads per user,
so if you only have a few concurrent users than all the extra cpus
won't help much, since geoserver's renderer doesn't use multiple cpus
per thread.
On Fri, Nov 26, 2010 at 4:02 PM, Ariel Nunez <[email protected]
<mailto:[email protected]>> wrote:
On Fri, Nov 26, 2010 at 3:47 PM, Gabriel Roldán
<[email protected] <mailto:[email protected]>> wrote:
> looks like it's only using ~500MB, and 1G for the OS buffers.
> How're you running tomcat/jetty wrt to heap allocation? Be sure
to pass
> -Xmx as appropriate to the java virtual machine and follow this
where
> applicable:
> <http://docs.geoserver.org/stable/en/user/production/index.html>
I have this in catalina.sh:
JAVA_OPTS="$JAVA_OPTS -Xmx756m -XX:MaxPermSize=378m
-XX:CompileCommandFile=/opt/tomcat/conf/hotspot_compiler"
And logging is set to PRODUCTION, but I have not installed the native
JAI libs yet.
> I would go for the 7G RAM/20 cores instance. But that's gut feeling
> since I really doubt geoserver is gonna need _that much_ memory and
> think it would appreciate the extra cores.
Yes, this is also what I would like to do. I guess I will give Tomcat
around 4 gigs and Postgis 1.5 gigs (I need to host at least two
GeoServer backends for two different GeoNodes)
Ariel.
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game by
optimizing for Intel(R) Graphics Technology. Get started today with the
Intel(R) Software Partner Program. Five $500 cash prizes are up for grabs.
http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Geoserver-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/geoserver-users
