Wireshark-dev: Re: [Wireshark-dev] Best algorithmic way to implement MPTCP sequence number spac
I managed to get the interval tree working with only a few
modifications to the core R/B tree.
I now would like to compute the sequence number inter arrival times
(multipath may provoke more out of order hence it is a metric of
interest). To do that I need to get access to the time arrival of a
specific frame (i.e., I know the frame number) from packet-tcp.c.
For instance I have a packet with Sequence Number (=SN) 100 in packet
5 and I know that SN 99 is in packet 2 or in packet 8 (out of order)
and I would like to compute the delta time of arrival between the 2
packets.
I found the frame_data_sequence_find() function but it expects a list
of frames as first parameter. Any idea how to retrieve that list from
packet-tcp.c ?
Regards
2015-10-02 18:11 GMT+02:00 Matt <mattator@xxxxxxxxx>:
> Hi,
>
> (Question is at the end, I start with an Multipath TCP introduction (MPTCP) ).
> I would be interested in adding MPTCP sequence number analysis to
> wireshark, similar to what is done with TCP but taking into account
> MPTCP specifics:
> - MPTCP sequence number is 64 bits.
> - MPTCP uses a global sequence number (SN) space and maps this global
> SN to a TCP SN. For instance, an MPTCP socket wants to send the bytes
> 1 to 10. It may send 6 bytes on TCP subflow 1 and the 4 other bytes on
> another subflow 2. On each subflow (=TCP connection), it will send a
> mapping (via a TCP option called DSS=Data Sequence Signaling) saying
> for instance
> on subflow 1 I map MPTCP global SN 1-6 to local subflow SN 21-26,
> on subflow 2 I map MPTCP global SN 7-10 to local TCP SN 60-63.
>
> Keeping track of those DSS options/mappings would help analyzing the
> behavior of an MPTCP connection:
> - does it duplicate some packets over different paths (<=> do some DSS
> overlap on different subflows?)
> - it would allow to generate an MPTCP RTT/application latency since
> striping packets over several paths is likely to increase the
> application latency
> - also allow to deduce if we've captured all subflows or if there are
> holes in the MPTCP sequence number (maybe one subflow was emitting on
> the wifi interface and we didn't capture it)
> etc...
>
> Now here is my problem: what datastructure should I use to achieve
> what I previously describe in a way that could be accepted upstream ?
> or to put it differently how to retrieve a list of ranges (MPTCP start
> SN and end SN) from another SN range ? The interval tree sounds like a
> good solution (http://www.geeksforgeeks.org/interval-tree/) but I
> wonder if there is any advice on an alternate solution ? or ways to
> build the tree efficiently (like disabling rotations and balancing the
> tree only at the end or stuff like that, I am not an algorithmic
> expert).
>
> In any case, I plan to add an option to turn this DSS analysis on/off.
>
> Regards
>
> Matt
