That's weird.  The patch should have given this:

Received (libpcap)      which is pcapStat.ps_recv
Less:Dropped (libpcap)  which is pcapStat.ps_drop
Gives: Given to ntop    which is calculated as
pcapStat.ps_recv-pcapStat.ps_drop

But, oh...

Total Received by ntop  is myGlobals.device[].ethernetPkts.value

s/b receivedPkts, which SHOULD equal the ps_ calc.

Then receivedPkts - droppedPkts (dropped by ntop due to buffer overflow)
should equal ethernetPkts.

ethernetPkts SHOULD = Unicast (616,107) + Broadcast (190) + Multicast (111)

So I need to tweak the reporting lines a smidge.

But those numbers are MUCH more reasonable (dropped of 325 out of 616K or
0.05%).

It would be interesting to see the stats from textinfo.html regarding
"Packet Queue"...

The dropped by libpcap - 37,068 of 653,801 or 5.7% still seems high.  But
that may be packets left in libpcap's queue that ntop didn't take in time -
without digging into the implementation of libpcap on the different
platforms, there's no way to tell what ps_drop really means.  Or it could be
due to a filter expression (-B "xxx" - I don't remember if you are using
one).


There are a few messages on the web of problems with the fxp driver - see,
for example
http://lists.freebsd.org/pipermail/freebsd-bugs/2003-April/000325.html and
http://news.gw.com/freebsd.stable/41069.  Ring any bells?


-----Burton




-----Original Message-----
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] Behalf Of David
Touitou
Sent: Friday, October 24, 2003 2:36 PM
To: [EMAIL PROTECTED]
Subject: Re: [Ntop] Dropped by the kernel




Burton M. Strauss III wrote:

> Try this patch - I'm working on it to rationalize the drop counts.

Thanks.

> It won't fix the drops (that's either NIC, libpcap or ntop simply not
> accepting them fast enough from libpcap), but it should make the stats
more
> rational in both the PR_ and info.html/textinfo.html.

Just compiled and launched it.

Web page is not the same (as supposed) and, while hitting "traffic"
again and again, the stats seams to be ok :
        Received (libpcap)      653,801
        Less: Dropped (libpcap) 37,068
        Gives: Given to ntop    616,733
        Total Received by ntop  616,408
        Gives: Total packets    616,408
        Unicast                 100.0%          616,107
        Broadcast               0.0%            190
        Multicast               0.0%            111

CPU usage is about 33%.
21.7% user,  0.0% nice, 11.2% system,  1.6% interrupt, 65.5% idle

Network traffic is arround 7500Mbps/1600Mbps.


Second answer about the test I started a few hours ago.
Here are the results in /var/log/messages after stopping ntop (with kill
-2 PID) :
STATS: 47,711,705 packets received by filter on fxp1
STATS: 295,578 packets dropped by kernel
STATS: 0 packets dropped by ntop

That's 0.62% of packet dropped by kernel as read in the log (not website).


David.

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

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

Reply via email to