[Kok-Yong Tan <ktan@xxxxxxxxxxxxxxxxxxx>]

I've come across this blog:  

<http://tadek.pietraszek.org/blog/2006/11/15/disabling-tcp-checksum-offloading-on-mac-osx-needed-for-pcap-based-programs/>

Basically, it tells me that playing with disabling TCP offloading would be a bad idea for MacOS X.  Hope this helps somebody.

On Sun, Mar 27, 2011, at 10:51, Kok-Yong Tan wrote:

It says that "Wireshark 1.2 and above disable IP, TCP and UDP checksum by default."  Does this mean that it disables checksum [validation] by default?  If so, note that I'm running Wireshark 1.4.3 with default settings on a MacOS X 10.5.8 system and I'm still seeing the TCP checksum errors.

On Sun, Mar 27, 2011, at 01:19, Martin Visser wrote:

In pretty much all case where you are seeing TCP checksum errors on a server, it will because of the various TCP offload features of the NIC / driver. If can capture on the wire (that is using port-mirroring on a switch) this will confirm this, or alternatively turn off those features on your server temporarily while testing.

See http://wiki.wireshark.org/CaptureSetup/Offloading for details

Regards, Martin

MartinVisser99@xxxxxxxxx

On 27 March 2011 10:07, Kok-Yong Tan <ktan@xxxxxxxxxxxxxxxxxxx> wrote:

On Sat, Mar 26, 2011, at 18:53, Kok-Yong Tan wrote:

Just for kicks, I decided to do a wireshark trace of AFP-over-TCP conversations between my Apple MacOS X 10.4.11 Tiger server and my Apple MacOS X 10.5.8 Leopard (PPC) client.  Surprisingly, I'm seeing lots of TCP checksum errors (no ssh going on here in the connection since it's all protected on my internal LAN) on packets going in both directions.  Now, if the TCP stack were damaged either on the client or the server or both, I would expect connection issues and all packets going through to exhibit the checksum errors.  But I don't and not all packets are exhibiting checksum errors between the two machines.  Only some.  Of course, this is manifesting itself in slower than expected throughput between the server and client since I assume that TCP checksum errors result in retransmits.  The server is connected to a ZyXEL GS2024 switch via LACP 802.3ad with 1 IP address in use in the two-NIC bonded pipe.  Could this be causing the TCP checksum errors?

More info on this:  I'm beginning to think that the LACP/IEEE802.3ad bonding of the server with the switch has nothing to do with it as I'm seeing the same checksum errors between the client (which only has one NIC and doesn't use LACP/IEEE802.3ad) and public servers hosted at akamai.net, doubleclick.com, etc., and even my externally hosted mail server.

--

Reality Artisans, Inc.             #   Network Wrangling and Delousing

P.O. Box 565, Gracie Station       #   Apple Certified Consultant

New York, NY 10028-0019            #   Apple Consultants Network member

<http://www.realityartisans.com>   #   Apple Developer Connection member

(212) 369-4876 (Voice)             #   My PGP public key can be found at <https://keyserver.pgp.com>

Attachment: smime.p7s
Description: S/MIME cryptographic signature