Date: Thu, 15 May 2003 14:16:46 -0500 From: Dan Nelson <dnelson@allantgroup.com> To: David Smithson <david@customfilmeffects.com> Cc: questions@freebsd.org Subject: Re: N! packets dropped by kernel Message-ID: <20030515191646.GK23782@dan.emsphone.com> In-Reply-To: <1053023794.23535.40.camel@blargh.customfilmeffects.com> References: <1053023794.23535.40.camel@blargh.customfilmeffects.com>
next in thread | previous in thread | raw e-mail | index | archive | help
In the last episode (May 15), David Smithson said: > Hi. I have a situation which may or may not be a problem. Here's my ip > configuration: > > nge0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> mtu 1500 > media: Ethernet autoselect (1000baseTX <full-duplex>) > nge1: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> mtu 1500 > media: Ethernet autoselect (1000baseTX <full-duplex>) > > I've been monitoring net traffic with tcpdump. Most traffic is SMB > and NMB. Tcpdump reports a very high dropped:received packet ratio. > For example, a few second of capture during peak traffic returns: > > 34964 packets received by filter > 34085 packets dropped by kernel > > Should I be concerned? I'll include full tcpdumps on both interfaces if > necessary. Thanks for your time. It means you need a faster CPU :) Tcpdump was only able to display half the packets it got, and the kernel had to drop the rest. Depending on what you're doing, writing to a file (-w logfile.txt), grabbing less bytes per packet (-s), limiting which packets to display (with a tighter filter expression), or raising the in-kernel buffersize (sysctl debug.bpf_bufsize) may work as well. I'm capturing (not decoding; just writing to disk) packets from four fxp interfaces on a 586-200 (no MMX even!), and my CPU load doesn't go over 10%. A machine 10x faster should be able to monitor two interfaces 10x the speed of mine with no problems. -- Dan Nelson dnelson@allantgroup.com
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20030515191646.GK23782>