Cluster I/O Protocols (CIP) Configuration and Management Manual (H06.16+, J06.05+)

Troubleshooting Network Applications Using Tcpdump
To trace and collect network traffic, you can use the Linux tcpdump command. On systems running
J06.06 or later J-series RVUs or H06.17 or later H-series RVUs, you can then transfer those files
to the NonStop host.
CAUTION: For pre-J06.06/H06.17 RVUs, do not use SFTP to transfer dumps and logs from
CLIMs to the NonStop host system. The only supported SFTP application for transferring files between
the CLIM and the console on pre-J06.06/H06.17 RVUs is PuTTY SFTP. Do not use PuTTY to enter
CLIM commands.
NOTE: Since tcpdump is a network-sensitive command, the provider option be specified to
CLIMCMD if the MULTIPROV option is ON. If the MULTIPROV attribute is set to ON, see the
Considerations section under prov(1p).
NOTE: In pre-J06.09/H06.20 RVUs, tcpdump can show large IP packets even when the jumbo
frame feature is not enabled on the CLIM interface. This can occur because Ethernet ports on the
IP and Telco CLIM were configured with TCP Segmentation Offloading (TSO) in hardware. TSO
enables the network stack to buffer a large stream of data (much larger than the supported MTU
of the medium) to the NIC. The NIC hardware segments it into MSS-sized packets with proper
sequence numbers. Because the packet-capture engine used by tcpdump operates above the
interface layer, the packet-capture engine captures the packet prior to the NIC-hardware
segmentation. This behavior results in packets captured by tcpdump showing a larger packet size
than the supported MTU. Packets passively captured on the wire (using a mirrored port on the
switch) would show the correct packet size.
Running tcpdump can impact system performance. These suggestions can reduce this impact:
1. Run tcpdump with -n to avoid name resolution.
2. Save the traces to a file with -w. This option causes the raw packets to be saved instead of
being parsed and printed in real time.
3. Do not run tcpdump with -i any.
4. Limit the capture size to the minimum when possible. Specifying -s 1500 can increase the
performance impact.
Running tcpdump
Here are steps for collecting traces by using tcpdump:
1. Create a trace directory on the CLIM
2. Identify all the TCP/IP attributes required to focus on only the data to be traced.
• Use the -n flag to avoid reverse look up of IP addresses
• See the tcpdump man page for details.
3. Capture the trace data and save to the trace file in bin format.
NOTE: You must be logged on as a super group user to run tcpdump.
4. Move the collected trace file to the NonStop host or the NonStop console.
5. Remove the trace file and trace directory from the CLIM.
6. Decode the trace file.
Example 15 Using tcpdump
1. Create a trace directory on the CLIM.
Use climcmd clim-name mkdir tracedir
> == Create a temporary directory on the CLIM
>
126 CIP Configuration and Management