So, I run this loadtest program on my work computer. It's going, I can see the components I'm testing registering events (via the realtime viewer [1] I wrote for syslogintr [2]). Everything is going fine … and … … then … … … t … h … e … … c … o … m … p … … u … … … t … … … … e … … … … … r … … … … … … … … s … … … … … … … … … … l … … … … … … … … … … … … … … … o … … … … … … … … … … … … w … … … … … … … … … … … … … … … … … … … … … … … … … … … … … … s …
It takes about ten minutes to type and run, but this:
>
```
[spc]saltmine:~>uptime
14:44:20 up 6 days, 23:12, 10 users, load average: 2320.45, 1277.98, 546.61
```
was quite amusing to see (usually the load average is 0). Perhaps it was just a tad ambitious to simulate 10,000 units on the work computer (each unit its own thread, running a Lua script [3]—yes, even after the modifications to the Lua interpreter [4]).
Also amusing was this:
>
```
[spc]saltmine:~>free
total used free shared buffers cached
Mem: 3910888 640868 3270020 0 35568 185848
-/+ buffers/cache: 419452 3491436
Swap: 11457532 544260 10913272
```
Yes, eleven gigabytes of memory were shoved out to the disk, so most of the slowless was due to thrashing [5].
Perhaps I should find some fellow cow-orker's computer to run this on …
[1] https://github.com/spc476/syslogintr/blob/master/realtime.lua
[2] https://github.com/spc476/syslogintr
[5] http://en.wikipedia.org/wiki/Thrashing_(computer_science)