Steve Dimoff wrote:

Hi.
Very unreliable measurement on a RH7.3 running as a virtual machine on a Xeon 2G4.


Has anyone got any speed comparison figures between 2.63/64 and 3?
Even rough anecdotal evidence would be very helpful.

And are there likely to be any major speed changes between the currect rc releases of 3 and the final "production" release?

2.63 (distrubution + bayes) 2449 spams, 120 falsenegs, average of 0.42 seconds/msg

3.0pre3 (distrubution + bayes)
2663 spams, 480 falsenegs, average of 0.18 seconds/msg

** Please note that 3.0 corpus was an expanded version of 2.63 corpus
** 3.0 run with Perl 5.8 compiled ad-hoc, 2.63 went with Perl 5.6 as of RH7.3




With some extra rules:

2.63 with bayes and SARE_ADULT, SARE_BAYES_POISON, ANTIDRUG, TW
2606 spams, 22 falsenegs, 0.90 seconds/msg

3.0pre3 with bayes and SARE_ADULT, SARE_RND, SARE_BAYES_POISON, SARE_OEM, ANTIDRUG, EVILN, TW
2663 spams, 5 falsenegs, average of 2.88 seconds/msg


** Please note that I was running tests in a virtual machine and I was more interested in FN/FP rate and hit ratios rather than performance at this stage. I have reasons to believe that 3.0's 2.88 s/msg were due to virtual machine being interfered by some other process on the host server.


HTH,BID'TTS Paolo


-- QRPp-I #707 + www.paolocravero.tk + I QRP #476 Beacon @ 28.3219 MHz + QRPp + QRSS3 + JN35TC

Reply via email to