From e4f42c701172b6b27dea2126e095e2526bcdb365 Mon Sep 17 00:00:00 2001 From: Pierrick Louin Date: Sat, 6 Mar 2021 01:23:56 +0000 Subject: NFVBENCH-190: Add a 'i40e_mixed' option, trex accepts other i40e driven ports to run in kernel mode on the same board Change-Id: I33b1d848f409e0431bd4e247456b1d697d8b9d04 Signed-off-by: Pierrick Louin --- nfvbench/cfg.default.yaml | 13 +++++++++++++ 1 file changed, 13 insertions(+) (limited to 'nfvbench/cfg.default.yaml') diff --git a/nfvbench/cfg.default.yaml b/nfvbench/cfg.default.yaml index 28f84f4..8e822e5 100644 --- a/nfvbench/cfg.default.yaml +++ b/nfvbench/cfg.default.yaml @@ -367,6 +367,19 @@ cache_size: 0 # The cache size is actually limited by the number of 64B mbufs configured in the trex platform configuration (see Trex manual 6.2.2. Memory section configuration) # Note that the resulting value is finally capped to 10000, whatever the requested size is (by design limitation). +# Specification of the TRex behaviour dealing with the i40e network card driver issue: Trex-528 +# see https://trex-tgn.cisco.com/youtrack/issue/trex-528 +# This issue states that if other ports, in the same card, +# are in kernel mode, they could impair traffic counting. +# Can be overridden by --i40e-mixed +# Values can be: +# ignore - don't consider the case (default) +# exit - should the case arise, exit (TRex default behaviour) +# unbind - unbind kernel bound ports (the former NFVbench behaviour) +# The 'ignore' option might be OK as soon as the issue has been fixed in the driver. +# The 'unbind' option should not be used! who knows the current use of other ports? +i40e_mixed: + # Trex will use 1 x 64B mbuf per pre-built cached packet, assuming 1 pre-built cached packet per flow, it means for very large number of flows, the number of configured mbuf_64 will need to be set accordingly. mbuf_64: -- cgit 1.2.3-korg