[VPP-537] vpp dropping ping frames when using option -s2003 (or above)

[vvalderrv]

Description

Testing jumbo frames between two VMs on two different compute hosts, I wasn't able to send frames over a bond sub-interface using values >= 2003.

Used 'trace add' and 'show trace' commands on receiving vpp, using -s2002 (or less) I see the icmp requests and response packets. Above 2002 I don't see any ICMP frames being captured. I did notice sending pings >= 2003 that the parent bond interface displayed, via 'show hard BondInterface0' command, that the 'rx missed' counter was incrementing.

Using:

vpp# sho ver

vpp v16.09 built by jenkins on centos-7-805 at Fri Sep 16 23:58:34 UTC 2016

Assignee

Andrew Yourtchenko

Reporter

Rich Curran

Comments

• florin.coras (Tue, 17 Oct 2017 14:31:13 +0000): Thanks, Andrew!
• ayourtch (Tue, 17 Oct 2017 09:35:05 +0000): 61459c9 made it such that the ping data size is limited by VLIB_BUFFER_DATA_SIZE, which is 2048. So I would expect this to still not work (as a limitation), but this time with an error message. So I have reclassified this as an improvement. The patches are welcome course!
• florin.coras (Tue, 17 Oct 2017 01:50:25 +0000): Andrew Yourtchenko, is this still an issue in 17.10?
• tiswanso (Mon, 5 Dec 2016 17:15:02 +0000): This is seen on Cisco UCS VIC 1227

(Rich and my environment)

Original issue: https://jira.fd.io/browse/VPP-537

[vvalderrv]

Thanks, Andrew!

[vvalderrv]

61459c9 made it such that the ping data size is limited by VLIB_BUFFER_DATA_SIZE, which is 2048. So I would expect this to still not work (as a limitation), but this time with an error message. So I have reclassified this as an improvement. The patches are welcome course!

[vvalderrv]

Andrew Yourtchenko, is this still an issue in 17.10?

[vvalderrv]

This is seen on Cisco UCS VIC 1227

(Rich and my environment)