<html>
<body>
<font size=3>At 09:42 AM 5/26/2005, Bernhard Fischer wrote:<br>
<blockquote type=cite class=cite cite="">On Thu, May 26, 2005 at
09:25:27AM -0700, Caitlin Bestler wrote:<br>
>The current OpenIB structure is, not surprisingly, tremendously<br>
>IB-centric. Any attempt to implement an iWARP RNIC through<br>
>them will have to go through several layers of obfuscation.<br>
><br>
>States do not match. Error codes do not match. Completion<br>
>Statuses do not match. There is a lot of information that<br>
Statuses? stati<br>
>does not translate to iWARP, and there is information that<br>
>is missing.<br>
such as.. ?<br>
><br>
>RNIC-PI is at least an attempt at providing full control over<br>
>both iWARP and IB while making as much common as<br>
>possible.<br>
Please point me to any infrastructure which allows any private peer
to<br>
steek at RNIC. At least at peer-level with a working driver which<br>
includes a working ethtool -C infrastructure..<br>
><br>
>I believe it is possible to define a verb set that is largely<br>
>compatible with current OpenIB IB verbs and have the<br>
>same transport neutrality that RNIC-PI has. But this will<br>
>not be just poking around and tweaking one or two <br>
>elements in the OpenIB code.<br>
[What is RNIC anyway? dict rnic returns just NIL, so what? Don't mean
to<br>
sound rude, but please provide at least a _little_ background for<br>
newcomers, will you?]</font></blockquote><br>
A RNIC is a defined term in the RDMA Consortium specifications which were
the basis for the IETF wire protocols. The specification and
associated presentations explaining all of this is reasonable detail are
available at the rdmaconsortium web site. There is also a good
architectural block diagram at the OpenRDMA website that goes through the
software architecture.<br><br>
Mike<br><br>
<blockquote type=cite class=cite cite=""><font size=3>><br>
>Is there a real willingness in OpenIB to make a transport<br>
>neutral verb layer? Has anyone even *read* the RNIC-PI<br>
Not being involved (heared of \"community\" before?), still
yes.<br><br>
Care to point (once again, as i'm buzzword agnostic) to the<br>
relevant IEEE/whatever RFCs? TIA.<br>
>for substance rather than "pure Linux" debating
points?</font></blockquote></body>
</html>