Omnibus CPU-load
This is an objective, reproducable performance measurement.
It measures the impact of a number of events on the cpu-load of the objectserver.
Results
First test reflects a Steady state.
The number of events is 100 times increased in steps of 100.
Each event has a probability om maximum of 3, 5, 10, 25 %.
tests om machine T1
Vertical: CPU-load in percentage (0..100) of the nco_objserv process.
Horizontal: Number of events (0..10,000) defined in the simnet.def file (times 100): .
X-ax * 100
Based on this hardware, the load is lineair with the number of events until the amount of 5000 is reached.
All events have a chance of re-occurance of maximum 3%, 5%, 10% or 25%.
At a certain point, the CPU becomes 100% busy. Not just with the OBJSERVER, but also other tasks, like simnet-probe, require CPU-time.
After 7000 lines, the simnet probe uses more then 20% cpu load.
tests om machine T2
Vertical: CPU-load in percentage (0..100) of the nco_objserv process.
Horizontal: Number of events (0..10,000) defined in the simnet.def file (times 100): .
X-ax * 100
tests om machine T1 and T2
10,000 lines/ 20,000 events all with maximum prob of 3%
Vertical: CPU-load in percentage (0..100) of the nco_objserv process.
Horizontal: Number of events (0..10,000) defined in the simnet.def file (times 100): .
X-ax * 100
10,000 lines/ 20,000 events all with maximum prob of 5%
Vertical: CPU-load in percentage (0..100) of the nco_objserv process.
Horizontal: Number of events (0..10,000) defined in the simnet.def file (times 100): .
X-ax * 100
10,000 lines/ 20,000 events all with maximum prob of 10%
Vertical: CPU-load in percentage (0..100) of the nco_objserv process.
Horizontal: Number of events (0..10,000) defined in the simnet.def file (times 100): .
X-ax * 100
10,000 lines/ 20,000 events all with maximum prob of 25%
Vertical: CPU-load in percentage (0..100) of the nco_objserv process.
Horizontal: Number of events (0..10,000) defined in the simnet.def file (times 100): .
X-ax * 100
The Test
A script generates the simnet.def file. Each event has a random possibility chance, with a pre-defined maximum.
After some time, all events will have occured at least once after which de-duplication takes effect.
The cpu-load of the object-server process is measured a number of times. The highest and lowest values are discarded, and the other values are used to calculate the everage.
The number of events is incremented by 100, simnet-probe is restarted and after settle down period, the load is measured again.
After the file simnet.def contains 10,000 lines, the test stops.
In real life, the chance that a cetrain event occurs is very slim, however, considering the total number of possible events, the chance that some events occur is certainly not NULL.
Currently, all events have probabilities within a fixed range (1..10%). This reflects a stable situation, without bursts of events.
hardware
The test machine T1
- CPU: GenuineIntel, Mobile Pentium II
- Speed: 365.868 MHz
- Cache: 256 KB
- Mem: 63455232
- Swap: 69632000
- OS: RedHat Linux 7.3 "Valhalla"
- Kernel: 2.4.18-3
The test machine T2
- CPU: GenuineIntel, Pentium III (Katmai)
- Speed: 448.061 MHz
- Cache: 512 KB
- Mem: 129675264
- Swap: 263168000
- OS: RedHat Linux 7.3 "Valhalla"
- Kernel: 2.4.18-3