Why OMG limit submessage size in RTPS no more than 64k？

[wddpfx]

According to DDS standard ，the type of field submessageLength in structure of the RTPS Submessages is uint16，so submessage size is no more than 64K. But if there is process A and B in the same host, and I shut down UDP transport, so there is just SHM transport which is aimed to use large fragement size to reduce fragement count. But the type of submessage is uint16, so no matter what transport I use, the size of fragment is no more than 64k.
why?

[MiguelCompany]

The standard allows submessages larger than 64K under certain conditions. This is exploited by the SHM transport when maxMessageSize is greater than 64k

[wddpfx]

@MiguelCompany ,Yes， I have noticed this. But in the RTPS Submessages, the type of field fragmentSize in datafrag is ushort, The max fragment size is also 64K.
So what is the meaning of this word "This makes it possible to send Submessages larger than 64k".

And When I have to transport a data which size is 10M, I want to reduce fragment count and use large fragment size which is up to 1M. It seems I can not set the fragment is 1M.

[cferreiragonz]

Important part is:

In case submessageLength==0, the Submessage is the last Submessage in the Message and extends up to the
end of the Message.

This means that there won't be fragmentation, and the message size will be extended to the total size of the Submessage. In your case, there will be only one DATA msg of ~10 M, no DATA_FRAGs.

[wddpfx]

But in my case, if the maxMessageSize SHM transport allows is 1M, then the data size of my topic is 10M, Now how could there be no DATA_FRAGs?

[cferreiragonz]

Then yes, your DATA_FRAGs will be of 64K size, there is no way to use DATA_FRAGs of bigger size. Either you set the maxMessageSize to allocate one message and you send 10M messages, or you fragment your message in 64K DATA_FRAGs.

However, there can be certain alternatives to your use case like adding multiple DATA_FRAGs into one single RTPSMessage. Although this is not currently on our short-term roadmap, we encourage you to contact our support team at [email protected], which can help you to accelerate this feature or offer dedicated help tailored to your specific use case.

[wddpfx]

I want to say is it a fault about this DDS standard?, Why don`t fastdds modify the type of field fragmentSize in struct datafrag to uint32.

[fenggaobj]

Perhaps this RTPS submessage standard was designed to be compatible with both SHM and UDP transports simultaneously. Due to the limitations of the IP header protocol (where the total length field is of type uint16_t), the maximum allowable size for a UDP packet is 65,535 bytes. Consequently, the maximum segment size for DDS UDP transport is also set to 64K. As a result, SHM has also adopted this standard. However, this is just my personal understanding and is provided for reference only.

[wddpfx]

No，UDP transport has been limit the maxMessageSize is 65500.so there is no need to limit the size in submessage