How much RAM / Items you have?

[nibb]

I wonder how much RAM the server where you run Zabbix has and now much items you monitor.

I configured Zabbix for 256 MB ram since im testing it. Low resources on zabbix agents and servers was the reason i wanted it. But here i want to discuss the server resources and not the agents. I dont actually care how much the server requires if the agents stay low. Since the monitored machines are my main interested, the zabbix servers can be trown in ram and more power anytime.

My install just default, using passive checks. zabbix_agentd that would be.

I tuned the linux template to use 60 seconds on most items that had below 60 so its not so heavy on the server. Only the network /nic items i left below 1 minute. Other items i increased to 5 minues like check free space, etc.

After the second server monitored with the linux template it began to swap. I deactivated 10 items on the Linux template. Increased Ram to 300. I have 5 servers now monitoring with Zabbix and 1 web check (zabbix interface) and its swapping ram in disk at 50%. Im going to increase it to 350 MB ram now and so on to see when it stops swapping. What i figured out currently Zabbix requires more or less 75 MB ram per server. I have now items 440. The CPU is very powerfull im just limiting the ram.

I dont think it leaks memory but I want to know when adding Ram wil be enough so it stops swapping. I cant imagine if i monitor 50 servers. I would probably need 32 GB ram with this setup.

What I have noticed is that the web interface is the one hitting the server hard. When you are logged or load the webinterface the CPU and Ram increases allot on each single load. Im not sure if this my setup but when you leave the web interface it goes lower and its considerable what it goes down.

I also would appreciate what settings you guys tune in the mysql to fit zabbix more properly since besides mysql there isnt anything hitting the resources. All this without loading the web interface, since that ones hits it pretty hard as well. I suspect its mysql. Will tuned to see if this goes down.

[jobster]

I'm running zabbix on a Xeon E5120 (Dualcore 1.86GHZ) bladeserver with 8GB of ram, the database is using about 4-5GB of the ram, but my cpu is running quite hot at the moment. Might have somethint to do with the fact that it is monitoring 330 hosts and around 10500 items. out of those 330 hosts around 215 are snmp-devices (mainly switches and routers) the rest are agent-based (servers)

[nibb]

I'm running zabbix on a Xeon E5120 (Dualcore 1.86GHZ) bladeserver with 8GB of ram, the database is using about 4-5GB of the ram, but my cpu is running quite hot at the moment. Might have somethint to do with the fact that it is monitoring 330 hosts and around 10500 items. out of those 330 hosts around 215 are snmp-devices (mainly switches and routers) the rest are agent-based (servers)

Well, thats a far better ratio then my install. Thats between 15 MB ram per hosts and i was at 75 MB. Main problem was apache and mysql so i made a tiny apache config to put it below 100 MB ram and mysql i also made a tiny optimization. I go it under the 350 MB without swap but the server was left with 0 memory that you could not even reboot anymore. Finally its a 400 MB ram, still with the tiny configs. Wondering if apache can be replaced by lighthttpd or something not so ram intensive since I only need the web interface to server 2 or 3 connections only.

Maybe the mysql and apache starts working correctly over 1 GB and I can get to your specs but making the match in hosts and items i have I would not get to the halfs of servers this way.

[krimson]

You should be able to offload some by putting up a proxyserver or 2 if you really hit a limit. Also, 60s refresh intervals is still way too often imo. Sure it makes for nice graphs, but it will also stuff your database in no-time and consume unnecessary cpu and memory. Also I understand you run the zabbix server in a VM, people on this forum have proven this is not a good idea.

I use 300s intervals for all my items, or more for static items like agent version and the likes. I'm running on a single dualcore cpu with 8GB RAM, database running on a different server ofcourse. About 150 hosts. I do not see myself running into trouble any time soon

[nibb]

Im not running 60 for everything. Version check and other stuff like free space check its at 5 to minutes.

60 is the lowest i run.

About the VM its not a virtuozzo vm that is just a chrooted jail sharing kernel. Its a xen Vm. Xen emulates exactly a normal server, with his own kernel, you install the CentOS with the normal CDs and go a normal install. Its uses Swap, it has dedicated allocated ram not shared. Same for disk. It runs just like a server. Actually I never saw a software that worked under normal Linux and would not work on Xen since the emulation is exactly like a normal server. Vms are not so recommended because of disk I/O if you run several of them together.

I dont think it has anything to do with the VM but its just ram. CPU like I said stays below 0.1 and now that i increased ram it works well.

What I have figured out is this. Below 500 MB ram you will have trouble regarless if you have 1 server monitored of 20. Because Apache and mysql require allot or memory just to run. If you have enough ram for apache and mysql to run correctly there rest is all for your zabbix server and monitoring.

Thats the only explanation I have found since some people run 1 GB to 2 GB and monitor a dozen of hosts and on my server i saw that actually the ones that where consuming all ram where apache and mysql, zabbix servers consumed almost nothing compared to them.

So now that I tweaked apache and mysql to run below 250 MB its working correcly and I increased the ram to 500 MB and will be slowly increasing since i add more hosts. I wonder how much i have to increase for each server. 60 seconds maybe is to much but in total i run 400 checks, and 30% are 60 seconds checks. I saw other people running over 1000 thousands checks with 1 GB ram.