<tt>"Tommy Mogensen" <tommy.mogensen@uni-c.dk> schrieb am 01.12.2009 16:14:39: > I am looking for a free system able to monitor 3000-5000 hosts (Mainly > cisco routers, switches and ap's) via ping and snmp. I would prefer to > run everything on one server if possible. > > I have received a good deal on a machine with 16G ram, 2xSSD-disks > (maybe in raid-0), and 2xIntel E5520-CPUs. I would appreciate your input > regarding the performance issues should I use Nagios for this system. > > Is this configuration powerful enough, what are the limiting parts of > the setup and are some of the pieces unnecessary? I.e. I could go for > one cpu (4 cores) if nagios does not support threading to 8 cores....or > is the main bottleneck that I do not run it distributed? </tt> <tt>It depends on how you want to implement the monitoring.</tt> <tt>First of all you need to break down the services on a per minute</tt> <tt>basis. The number of checks does not matter, but in which timeframe</tt> <tt>matters a lot. If you plan on running 5.000 checks with 5 minute</tt> <tt>interval - all is cool. If you are running 5.000 checks with 1</tt> <tt>minute interval - you will need to tweak your nagios server a lot.</tt> <tt>If you are just using nagios for pure monitoring, then you have</tt> <tt>more then enough power.</tt> <tt>If you are looking for using ndo for visualisation with NagVis for</tt> <tt>example and/or graphing performance data with PNP4Nagios, then I/O is</tt> <tt>your biggest obstacle.</tt> <tt>Using SSDs for a DB is a two-fold sword - they runlike hell, but</tt> <tt>SSDs melt in hell ;) It really depends on the quality of the SSDs.</tt> <tt>If they are enterprise drives, they should be good to go for 1-2</tt> <tt>years until they drop dead. If they are midline drives, you will burn</tt> <tt>them quite fast.</tt> <tt>The most important thing to have for a nagios installation with</tt> <tt>local NDO and/or PNP4Nagios togeter with massive services is a</tt> <tt>BBWC (battery backed write cache) of at least 512MB to aggregate</tt> <tt>the written blocks and put it on 100% write / 0% read cache.</tt> <tt>This speeds up NDO/PNP like tenfold. Read performance is neglectable</tt> <tt>for nagios, write performance is all that matters.</tt> <tt>I'm running 2.000+ service checks per minute on 5GB, 2x 2GHz QuadCore</tt> <tt>machine with 2 local 10K hdd's in RAID1 with a local mysql DB for</tt> <tt>NDO, performance graphing of around 1.200 services with PNP4Nagios</tt> <tt>and hosting NagVis for visualisation via mysql on the same server.</tt> <tt>Average load over 1 month is ~2.7 with peaks going up to load 8 aprox.</tt> <tt>Before the use of a BBWC I had an average load of 8 with peaks</tt> <tt>around 20 due to I/O retention.</tt> <tt>Regards</tt> <tt> Sascha</tt> GFKL Financial Services AG Vorstand: Dr. Peter Jänsch (Vors.), Jürgen Baltes, Dr. Tom Haverkamp Vorsitzender des Aufsichtsrats: Dr. Georg F. Thoma Sitz: Limbecker Platz 1, 45127 Essen, Amtsgericht Essen, HRB 13522