Kanevsky, Arkady wrote:
It may be benefitial to split the range of SID for one TCP port into
several.
That way the all "base TCP ports equivalent" will have the same
mapping.
Mapping to support SDP will use SDP assigned Service IDs.
And IPoIB will have its own range similar to SDP for TCP ports.
The RDMA "native" will use the "base" TCP ports.
We can also add more ports for each of these categeories
to support multiple SIDs. But we as well
can use portmapper for them.
But transparent adoptable clients that support all 3 (or more of the
methods)
will have to try each of the assigned Service IDs using appropriate
transport
to see if server supports it.
Arkady
An application that wants to connect to destination port 53 is expecting a
particular application. The benefits of letting the application connection to
different applications depending on a transport that may be selected by some
underlying software is questionable to me.
If we're going to define a protocol that passes TCP/IP addresses in private
data, then the addresses should behave as close to TCP port numbers as possible.
Two CM REQs that contain the same address should expect to reach the same
destination.
- Sean
_______________________________________________
openib-general mailing list
[email protected]
http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
