You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
At Netdev 0x19, there were some discussions about hardware GRO: Eric
reported that when comparing performances of TCP with MPTCP single-path,
a performance drop can be explained by hardware GRO not being used with
MPTCP.
Apparently, some NIC's skip hardware GRO in presence of "unknown" TCP
options: anything but TCP timestamps.
It then seems interesting to write this down in the FAQ to help people
understanding why hardware GRO might not be used with MPTCP. While at
it, explain that software GRO works when TSO was used on the sender
side.
Co-authored-by: Mat Martineau <[email protected]>
Signed-off-by: Matthieu Baerts (NGI0) <[email protected]>
0 commit comments