Re: Load Balancing
Manny Gonzalez <gonzalu <at> nyp.org>
2005-09-03 15:49:36 GMT
I think that an NFS solution or SMB would be better for this. Or a san/nas
If you have let's say, AIX on IBM with a NAS, you can mount the same directory
on multiple boxes.
Now, on each box, collect a different part of the tree while pointing it to the
Using NFS is very similar, but this requires one of the boxes to be a master
repository for the data and a NFS center of the world. Using the NAS makes the
NAS the center of the world, not the individual polling hosts.
Another solution would be to have totally separate and distinct polling systems
with their own responsibilities... say:
HOSTA --- Switches
HOSTB --- Routers
HOSTC --- Perfmon
HOSTD --- Latency
HOSTE --- Miscellaneous
Each system would look like an independent Cricket box complete with its own
HOSTF --- Master Web Server
On this host, you will get creative with HTML and basically it is a master
linker to the individual poller systems. When someone clicks Routers on a master
page, they get re-routed to the HOSTB poller... etc. etc.
There are definitely more ways tha one to get this working. It will be nice when
it is more integrated in Cricket itself
John Clinton wrote:
> Hi Dermot,
> I assume the load balancing needs you have are on the poller side, not
> the web presentation. Correct me if I am wrong.
> I distributed our polling needs over several cricket boxes (pollers) by
> doing the following.
> 1. Setup cricket on remote pollers as usual.
> 2. Start polling on the remote poller, make sure rrds look good.
> 3. On the central/master Cricket box (where your users access
> grapher.cgi). Copy cricket-config/<whatever you setup> from remote
> poller to master cricket/cricket-config/. As you can see, the tree
> structure/nodes you poll need to be unique across all your cricket
> 6. Edit the remote cricket Target files on the master cricket server
> and add "collect = false" to every entry. If you setup custom Defaults
> on remote poller be sure to import the settings into the master cricket
> Defaults files.
> 7. compile on master cricket
> 8. create appropriate cricket-data directories for remote poller
> devices on master cricket box cricket-data/.
> 9. copy rrd files from remote poller to master poller in some routine
> fashion. Need not bother with *meta files.
> Took me a little while to figure this out since there was no
> documentation on it, but it seems to work quite well.
> John Clinton
> On Fri, 2005-09-02 at 11:14 +0100, Dermot Williams wrote:
>>Does anyone have any experiences with trying to load-balance Cricket
>>that they would care to share? We have reached a point with our
>>implementation where we are graphing nearly 9000 nodes. As a result,
>>the server that we are running Cricket on has started to creak at the
>>seams – some of our subtree-sets have grown so large that they were
>>taking over 20 minutes to run and weren’t graphing properly! I split
>>them out so they were smaller, but obviously running this number of
>>collector processes (plus web services etc.) is just hammering the box
>>Has anyone any ideas on how to implement a load-balance solution?
>>Also, anyone had any success solving the core dump issues with Perl
>>5.8.7 on FreeBSD?
Manny Gonzalez, CCIE #9013 .............. Network Design Manager
CORE Resources ................... NewYork Presbyterian Hospital
PH18-129 .................... Columbia University Medical Center
manny <at> nyp.org ............................... mag58 <at> columbia.edu
SF.Net email is Sponsored by the Better Software Conference & EXPO
September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf