Wireshark mailing list archives

Re: Experiencing Packet Loss in High Volume Packet Capture Application using DUMPCAP


From: Jeff Morriss <jeff.morriss.ws () gmail com>
Date: Mon, 26 Nov 2012 15:49:06 -0500

John Powell wrote:
Hi Everyone,

I am running CentOS 6.3 on a HP 8200 using 3TB WD Green drives using a EXT4 file system.

I am using Wireshark 1.8.2 compiled from source.

I am using DUMPCAP to rotate and store historical Packet Captures.

Whether I capture the packets with Wireshark or view the DUMPCAP created file, I see dropouts in the packets being captured.

I tried to turning off journalling but this did not seem to help much:

umount /dev/mapper/VolGroup00-LogVol_Data

/sbin/tune2fs -o journal_data_writeback /dev/mapper/VolGroup00-LogVol_Data

/sbin/tune2fs -O ^has_journal /dev/mapper/VolGroup00-LogVol_Data

/sbin/e2fsck -f /dev/mapper/VolGroup00-LogVol_Data


I have a attached a couple of IOGraphs from Wireshark showing the packet drops.

(Note that Microsoft documents aren't the most portable way of sharing... Many of us don't natively have a way to open them. Fortunately, Google frequently can...)

The document indicates that your disks are 71% busy writing about 38 Mbytes/sec and that you're periodically getting periods where almost *nothing* is captured and that those periods can be quite long (in one case it looks like about 500 msec).

In my mind, you're crossing into the territory where a dedicated capture device (which has been engineered for this kind of high-speed capture) is needed. You may be able to make some progress but you'll be reinventing a wheel that's already been solved (probably with much effort) by several vendors.
___________________________________________________________________________
Sent via:    Wireshark-users mailing list <wireshark-users () wireshark org>
Archives:    http://www.wireshark.org/lists/wireshark-users
Unsubscribe: https://wireshark.org/mailman/options/wireshark-users
            mailto:wireshark-users-request () wireshark org?subject=unsubscribe


Current thread: