Luca Deri wrote:

Hi all,

Hi Luca & all,


- use *BSD and enable polling

I just did, same server (FreeBSD 4.9-RC, P3-733) as yesterday.


Current bandwidth is arround 5Mbps (no udp game packets right now), I added device_polling option in kernel and HZ=1000.
With kern.polling.enable=0 :
ntop CPU usage arround 9%
arround 6.5% packets drop (from web interface)
With kern.polling.enable=1 :
ntop CPU usage about 9%
arround 6.5% packets drop (from web interface)


I tried both to change kern.polling.enable with ntop running and to change it with ntop stopped.

Under such "low" network load, device polling does not help.

Here are the packets size stats from the web interface :
        < 64 bytes   45.4%   2,589,400
        < 128 bytes  29.0%   1,654,627
        < 256 bytes  19.8%   1,127,276

Still the same unconsistancy between web interface and ntop log.
Web Interface :
        Total   5,703,202
        Dropped by the kernel   376,074 [6.59 %]
ntop log :
        STATS: 697,415 packets received by filter on fxp1
        STATS: 5,590 packets dropped by kernel
        STATS: 0 packets dropped by ntop

Why such difference ????

Changing buffer values (kern.ipc.maxsockbuf, net.inet.tcp.sendspace and net.inet.tcp.recvspace) does not change the CPU usage or packets drop percentage.

Anyone willing to help me or at least provide me a testfield for testing at Gbit speeds?

I don't have gigabit, but I have "not working" less-than-10Mbps 8-))


David


_______________________________________________ Ntop mailing list [EMAIL PROTECTED] http://listgateway.unipi.it/mailman/listinfo/ntop

Reply via email to