David Brown wrote:
Hello Steve, I am subscribed to various MLs such as Ant and Tomcat because I seem to find something that grabs my interest almost daily. And, the smartfrog link below is no exception. I went directly to smartfrog and I have been reading avidly for several hours.
And, I agree with the rhetoric so far that smartfrog is not arcane but is overwhelming to the new user because of all the features (reminds me of JMeter).
My current gig is a Tomcat/JBoss assessment for a company that is in dire need of performance tuning and some type of vertical of horizontal scaling
>such that their current web app and web service installations can
scale up to 25k users. I am currently using JMeter for remote
distributed testing
> but what I need more are multiple instances of their servlet
containers (Tomcat & JBoss). JBoss lends it self to multiple instances
very handily
>but their current Tomcat 5.5 by what I have seen on the Tomcat ML can
be problematic to configure and difficult to maintain as a
>stable multiple instance server. My question is (before I spend a
gazillion hours working-out the smartfrog examples) can smartfrog help
> toward creating a computing grid using disparate machines and
disparate systems of disparate JDKs/JREs and disparate Tomcat versions?
disparate JDKs is trouble. What we like to do in that world is push out
the right JDK versions, usually by uploading and installing the RPMs.
That said, as long as you are running java5+, SmartFrog is happy.
The way to view SmartFrog is the components and the runtimes to do large
cluster systems, but not some nice shrink-wrapped tool to do it for you.
What we use for big systems is:
- Anubis ( http://wiki.smartfrog.org/wiki/display/sf/Anubis ) so that
deployed nodes can find each other without a central manager. This is a
tuple space that notifies peers on a LAN when nodes come and go.
-a special anubisdeployer that deploys work to any machine matching the
requirements
-a CpuMonitor component that runs vmstat to determine current system
load. This is used to trigger requests new machines, and for machines to
declare themselves idle, when they can be returned to the pool.
There's a big dynamicwebserver example that installs and runs apache
httpd on demand; Supporting Tomcat should be similar; its just not (yet)
something anyone's put together
>What are the advantages of smartfrog over just using JMeter and
load-balancing? I read the PDF whitepaper: Globus Toolkit and smartfrog
but that reading did not lead to anymore confidence.
> What happened to the day when I had time to read the Loughran/Hatcher
Ant book and then go-to-work? Please advise, David.
I dont know what happened to time either.
Where things get interesting over basic load balancing is when you
allocate real/virtual machines on demand. With virtualisation (I'm doing
stuff with amazon's EC2 farm right now), you really can do interesting
stuff. There's some VMWare components being done in the Open Source
repository; demand allocation of VMs followed by configuration of the
deployed machines. This is both cost effective and agile.
see also:
http://people.apache.org/~stevel/slides/farms_fabrics_and_clouds.pdf
What I'd recommend you do is
1. grab today's release, preferably the self-installing JAR or the
linux RPMs
http://sourceforge.net/project/showfiles.php?group_id=87384&package_id=108447&release_id=576780
2. Get on the smartfrog-users mailing list and discuss what you want
to do; see if we can't come up with the right combination of things:
http://sourceforge.net/mailarchive/forum.php?forum_name=smartfrog-users
3. Pick something small, like database+app server, before going for the
200+ node server farms.
-steve
--
Steve Loughran http://www.1060.org/blogxter/publish/5
Author: Ant in Action http://antbook.org/
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
