[Linux-cluster] GFS and server performance = Application

[isplist at logicore.net]

> Are the same nodes asked to access individual subsets of the application
> paths or are all nodes handling everything?

At the top is an LVS server doing load balancing for the web servers. 

The web pages are on a shared GFS storage device.
The clustered web servers all have their /var/www/ mapped to the GFS partition 
where all of the web pages reside. 
This means that all web servers are sharing the web sites/pages so all are 
serving up the same copies of pages. There is zero load right now since it is 
not public. 

> But the point I was making was that if
> your test waits for a response before asking again, you may find that the
> throughput goes right down because latency goes up. If you are issuing
> 10 requests in parallel, that may cover up the latency increase.

Right and this is why I'm suspecting joomla because other applications seem to 
bring up good performance based on the same testing at least. I've tried 
various applications all the way from just an index.html page to a non Mysql 
app to a Mysql app to joomla. Joomla is the only one which is giving me these 
very low 2/4 requests per second results.

> Mounted filesystems that you aren't accessing won't be affecting the
> performance. It's pretty safe to ignore those for now.

Good to know.

> It would be worth
> looking into how Joomla handles it's file accesses. If it's constantly
> opening and closing lots of files for r/w access, that may well explain
> the 10x slowdown you're seeing.

Good point and I am posting a message in the joomla forums right now to get 
some more input on this. Another list member just posted some good links to 
GFS testing and fine tuning also which I'll take a peek at.

Mike