Server with Zabbix 1.0 installed (RedHat Linux 8.0, kernel 2.4.18-14, MySQL/MyISAM 3.23.54a-4, Pentium IV 1.5Ghz, 256Mb, IDE) is able to collect more than 200 parameters per second from servers being monitored (assuming no network delays).
How many servers can be monitored by Zabbix on the hardware, one may ask? It depends on number of monitored parameters and how often Zabbix should acquire these parameters. Suppose, each server you monitor has ten parameters to watch for. You want to update these parameters once in 30 seconds. Doing simple calculation, we see that Zabbix is able to handle 600 servers (or 6000 checks). In case if these parameters need to be updated once in a minute, the hardware configuration will be able to handle 600×2=1200 servers. These calculations made in assumption that all monitored values are retrieved as soon as required (latency is 0). If this is not a requirement, then number of monitored servers can be increased even up to 5x-10x times.
It is very important to have Zabbix system properly tuned for maximum performance.
General advices on hardware:
Many parameters may be tuned to get optimal performance.
General rule - keep value of this parameter as low as possible. Every additional instance of zabbix_server adds known overhead, in the same time, parallelism is increased. Optimal number of instances is achieved when queue, on average, contains minimum number of parameters (ideally, 0 at any given moment). This value can be monitored by using internal check zabbix[queue].
Optimal value is 3.
MySQL only. It is recommended to use DBSocket for connection to the database. That is the fastest and the most secure way.
This is probably most important part of Zabbix tuning. Zabbix heavily depends on availability and performance of database engine.
Example: why use system[procload],system[procload5] andsystem[procload15] if system[procload] contains all.