Using Apache Benchmark (ab)
$ ab -n 2000 -c 100 http://host
- 2000 requests to perform
- 100 multiple requests to perform at a time
$ w 17:28:41 up 17:10, 3 users, load average: 107,34, 71,08, 32,14 USER TTY FROM LOGIN@ IDLE JCPU PCPU WHAT
The result was a load average of 107.34 – on a Intel Atom Dual Core with HT (Hyper Threading). That was - until there - definitely my record. Although the optimal load of this system is a average of 4, the system was even accessible and able to reply other request (e.g. Cyrus, NFS, DNS, SSH connections).
$ stress -i 10000
…but just on
sync() to keep the CPU usage down.
On the left you can see the result of such a stress test and below an output of
w before the screenshot was taken.
$ w 17:28:41 up 22 min, 3 users, load average: 983,11, 869,82, 904,16 USER TTY FROM LOGIN@ IDLE JCPU PCPU WHAT
While the system reached a load average of over 1000 it was even reachable! So Munin (on another server) was able to connect to the Munin-Node and generate this graph: