[Xymon] Xymon performance scaling problems for Solaris zones?

[Andy Smith]

Mills, David (IS) wrote:
> Hi, all...
> 
> We've been running Xymon (4.3.3 -- yes, old, old, old...) very 
> successfully for the last several years in our environment, but have run 
> into a performance snag recently.
> 
> We run Xymon on all our Solaris zones -- up to 10 zones + the global 
> zone, at a time. Recently we've started configuring new Oracle LDOMS (64 
> x SPARC T5-8 processors / LDOM) with > 20 Solaris zones / LDOM and we're 
> seeing that after about 10 zones, Xymon starts to hog CPU in a 
> non-linear way, making the zones unusable.
> 
> Specifically, it's not any of the "Xymon" daemons, but it's all the 
> concurrent "vmstat" and "iostats" that are racking up the CPU cycles.
> 
> We understand we have several options available including instantiating 
> processor sets and pools, ulimits, etc, but we wanted to know if:
> 
> a) Anyone else has had to navigate this problem with such a large number 
> of concurrent *stat commands
> 
> b) Is it possible / advisable to monitor zones via Xymon using only a 
> single instance of Xymon running in the global zone, or possibly a 
> representative local zone?
> 
> The LDOMs all run Solaris 11, FWIW.
> 
> TIA for any tips!
> 
> david
> 
> ~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~
> David Mills
> Systems Administrator
> Northrop Grumman
> (512) 873-6665
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Xymon mailing list
> Xymon at xymon.com
> http://lists.xymon.com/mailman/listinfo/xymon

Dont know if this is going to help much, but we have a bunch of T4s 
running up to 22 LDOMs, each has Xymon (4.3 series) installed and the 
primary shows a load average peaking at 0.2 over the last 48 hours.  We 
also have T4s running zones, but the largest number of zones configured 
is 5, again, even a non-production global zone (where theoretically 
there is no application load at this time) load average sits around 1.0.
-- 
Andy