On Tue, Feb 6, 2018 at 9:07 AM, S. Jacobi <sjacobi@xxxxxxxxxxxxxxxx> wrote:
> On Tue, 6 Feb 2018 09:05:14 -0800
> Richard Sharpe <realrichardsharpe@xxxxxxxxx> wrote:
>
>> On Tue, Feb 6, 2018 at 8:39 AM, S. Jacobi <sjacobi@xxxxxxxxxxxxxxxx>
>> wrote:
>> > We have a sender who send packets, each one gets a 16bit number.
>> > This number, I will call it packet ID, is strictly ascending, but
>> > starts again from zero if the 16bit range is reached.
>> > Then, the sender distributes the packet on multiple interfaces and
>> > we cannot make any assumptions how this is done. Packet IDs can
>> > appear arbitrarily on the interfaces, packet IDs can be reordered
>> > (although only in a very limited range), and packets need not be
>> > (and in fact are not) evenly divided onto the interfaces.
>> > On the receiving
>> > Our own capturing tool is rather simple. It spawns a thread for each
>> > interface, and the thread functions tries to read and process each
>> > incoming packet as fast as possible. This leads to the problem that
>> > if one interface receives more packets, the packet IDs read from
>> > different interfaces drift further apart, even going one full
>> > circle and so on and on.
>> > However, if we use tshark to capture from all interfaces and save
>> > the output to a file, the process this file with our tool,
>> > everything works fine.
>> > So, tshark needs to have some sort of synchronisation mechanism, to
>> > fairly distribute the reads from each interface. The packet
>> > timestamps in the capture file are not always ascending, there are
>> > a few jumps in it.
>> > I wasn't able to spot this mechanism in the code, so I'm grateful
>> > for any information on this.
>>
>> As far as I am aware it is the kernel that is doing this. Also, I
>> believe that only Linux supports the any device.
>>
>
> We are on Linux, yes, but we don't capture from any. tshark allows to
> specify multiple interfaces.
I have not looked at the code, but I suspect that it is something like this:
https://stackoverflow.com/questions/37294540/binding-the-af-packet-socket-to-all-interfaces
That is, the kernel is doing it.
--
Regards,
Richard Sharpe
(何以解憂？唯有杜康。--曹操)
