[Openib-windows] RE: IBDM and IBMgtSim Proposal Comments

[Eitan Zahavi]

Hi Hal,

> Where does that leave the OpenIB diagnostics and other applications
> which do not use this layer ? Does the simulator support these
> applications ?
[EZ] The simulator is a separate process that is a TCP/IP server.
Any application that will send and receive simulator messages (as defined by
simmsg.h) can connect to it. 
The osmv_sim (osm vendor layer for simulator) is just an example for such
client.
But one can even write perl or TCL applications that will connect. 
Actually the simulator provides a library named ibmscli that provides all
the required TCP/IP (client/server) logic and exposes a simple API for such
client.
Please see ibms_client_api.h for this API definition.  
> 
> In the OpenIB architecture, umad is the lowest layer library and the
> diagnostics are built on that. The OSM vendor layer is built on top of
> this. So when the umad and mad libraries are ported to Windows,
> everything on top of this will work. This includes all diagnostics
> (OpenIB ones as well as the additional tools you are proposing to add).
[EZ] There is no need to port UMAD to windows!!! We already have OSM Vendor
ported to it. It works on top of the existing IBAL API (actually this is the
first OSM Vendor that was ever built).
> 
> Also, it appears that the Mellanox simulator is builtin to each
> simulated process and so each instance is separate which limits its
> usability.
[EZ] This is totally incorrect - as explained before. We were able to
simulate OpenSM, osmtest and IBIS (IB Inband Services - which is a simple
API for sending MADs) all connected and communicating over the simulator. We
were also able to simulate multiple SM's handoff processes. Where did you
get this idea from?
> 
> In the Voltaire architecture proposed to be moved forward to OpenIB, the
> simulator is standalone and hence is useable by all applications in the
> network which can be on different nodes (TCP connections): OpenSM (as
> many as you want in the subnet), diagnostics, SA clients, etc. Also, the
> Voltaire simulator architecture can use an input file recorded from a
> real IB fabric where it appears that this would be manual in the
> Mellanox scheme.
[EZ] The Voltaire simulator was never made public and actually has no
benefit over what is provided by the IBMgtSim. 
> 
> The file formats (netlist and topology) are for further study. There are
> some good ideas in these file formats which would be incorporated into
> the ibnetdiscover format.
[EZ] Thanks.
> 
> -- Hal
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/ofw/attachments/20050707/353224f8/attachment.html>