<html>
<body>
<font size=3>At 04:57 PM 10/14/2004, Fab Tillier wrote:<br>
<blockquote type=cite class=cite cite="">> From: Michael Krause
[<a href="mailto:krause@cup.hp.com" eudora="autourl">mailto:krause@cup.hp.com</a>]<br>
> Sent: Thursday, October 14, 2004 4:21 PM<br>
> <br>
> Why not just leverage the SDP port mapper protocol already defined
in the<br>
> RDMAC version and avoid having to provide a new address
family? The port<br>
> mapper protocol is interconnect independent and will enable
sockets<br>
> applications to more easily be executed transparently. It
seems<br>
> counterproductive to continue to pursue a new address family.<br>
> <br>
> BTW, the new port mapper protocol will also work with the new
async<br>
> sockets and memory management API that is nearly complete (should
be<br>
> approved soon within the OpenGroup). This would greatly
enhance socket<br>
> application design and provide greater performance when operating
over a<br>
> RDMA interconnect than traditional BSD sockets.<br>
> <br><br>
A while ago, there was some discussion of having transparent port mapping
be<br>
a bad thing, and a security vulnerability of some sort. Note that I
don't<br>
personally believe that.</blockquote><br>
The SDP spec submitted to the IETF as a draft has the port mapping.
The HP IETF gurus did not have a problem with the port mapping.
Discussions with various IETF AD did not generate a negative reaction but
one never knows how the IETF will act. In general, the exchange has
measures to mitigate any DOS attacks and conserve resources. Take a
look and please consider as this is the best way to get a large number of
sockets applications to quickly operate over RDMA interconnects without
requiring any source code changes. <br><br>
Mike</font></body>
</html>