As the plan is to embedd python as scripting language into the honeypot, I ran a benchmark on a testsuite. The 'testsuite' is a c core which accepts connections, and allows python to deal with the input. The protocol used for benchmarking is http, the service serves a non static html page.
To benchmark, I ran the apache benchmark tool ab
ab -n 100000 -c 15 http://localhost:8080/bar
|Document Length:||191 bytes||191 bytes||191 bytes|
|Time taken for tests: (seconds)||28.47||102.15||31.8|
|Total transferred: (bytes)||38600386||38200000||37600376|
|HTML transferred: (bytes)||19100191||19100000||19100191|
|Requests per second: [#sec] (mean)||3512.04||978.95||3144.45|
|Time per request: [ms] (mean)||4.27||15.32||4.77|
|Time per request: [ms] (mean, across all concurrent requests)||0.29||1.02||0.32|
|Transfer rate: [Kbytes/sec] received||1323.89||365.19||1154.61|
Python 3.0.1 serves about 978 requests per second, where Python 2.6.2 does 3512 requests/second.
|Percentage of the requests served within a certain time (ms)||2.6.2||3.0.1||3.1a|
|100% (longest request)||3004||3019||3006|
Obviously Python 3.0 does something wrong, talking about factor 4x, it is noticeably slower. But I'm glad to see it is fixed in 3.1 already.
I tried to benchmark using unladen-swallow, the llvm based python project run by google, but was not able to link the code properly. (undefined symbol: _ZTIN4llvm14ModuleProviderE)
As Python 3.1a is almost as fast as 2.6 for our needs, and it can be tricky to port python 2.6 code to 3.x, we will use python 3 as embedded scripting language.