[multipathtcp] Some comments on RFC 6824

[Rao Shoaib <[e-mail]>]

Hi All,

I am implementing MPTCP in Solaris and have a few comments below based 
on my implementation experience so far.

1) Currently the RFC says that checksum in DSS option 'SHOULD' be 
present.  Requiring checksum is a performance killer because it has to 
be calculated in the software and needs to be calculated over the ENTIRE 
mapping. This also means that the receiver has to wait to receive the 
complete mapping worth of data before it can be processed. More 
importantly these days most  NIC's support HW checksum offload and that 
feature is commonly used, this requirement makes this important HW 
feature useless.

I am hoping that the requirement of requiring checksum in the DSS option 
be relaxed and may even not be required in a data center.

2) The RFC does not say what the initial subflow sequence number should 
be. Can this be made clear. Why not use the TCP sequence number of the 
subflow.

3) When calculating the HMAC, token, key's sequence number etc it is not 
at all clear which values have to be in network byte order and which 
not. Can this be explicitly specified in the RFC.

4) Can we not allow dropping a segment after it has been acked at the 
subflow level. It is very unlikely that a network element would drop a 
TCP segment after it has been processed and acked. Making this change 
would make buffer management much easier. If an element can drop a 
packet than it should not ack,  but other implementation should not be 
burdened by this requirement.

Thanks,

Shoaib