[Linux-cluster] Cluster/Server Monitoring tool

[isplist at logicore.net]

The two main processes I see at the top using the top linux tool is that  
gfs_scand and httpd are taking up most of the servers time.
I need to figure out why the servers are acting so sluggish when they don't 
have anything to do but serve up web pages. I also need to optimize them just 
for serving pages, LAMP style. 

For some reason, sitting idle, they are still always at the 0.50 load range, 
jumping to 1.00 and 1.50 constantly and no one on them. None of the other 
nodes are doing this, they are pretty much at 0.00, only the web servers are 
doing this. 

Oh, one of them is also a web server but serves up nothing but images.

Mike


On Mon, 19 Nov 2007 15:01:52 -0800, Lombard, David N wrote:
> On Mon, Nov 19, 2007 at 04:37:44PM -0600, isplist at logicore.net wrote:
> 
>> Ah, now that's a nice clear picture of what's going on. I looked at the
>> other
>> one (thanks for the lead) but it seems to be MRTG style output only?
>> 
>> Being able to put things into the database is something I like, so that I
>> can
>> go back and review and perhaps gain from a patern showing up.
>> 
> Don't let go of Ganglia so quickly, at least not for that reason.  Ganglia's
> data is placed into an RRDtool database, which allows for access; Ganglia
> is actually showing data from that DB.  You can also extract the data--the
> collected stats actually, not the raw data--for arbitrary access.
> 
> Another nice feature of Ganglia is that any node can glom the data from
> the multicast channel.