On 5/29/07, <b class="gmail_sendername">Bob Kossey</b> <<a href="mailto:bob.kossey@hp.com">bob.kossey@hp.com</a>> wrote:<div><span class="gmail_quote"></span><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
Another related question. Does OFED 1.2 now support multiple<br>independent IB fabrics<br>(multiple SMs, etc) connected to multiple HCAs on the same node? Are<br>there any<br>qualifications about which dimensions are supported with this, such as
<br>ipoib HA, SRP HA,<br>other types of failover, etc.?<br></blockquote></div><br>
Hi Bob, <br>
<br>
Generally speaking, as far as i am aware, by design the openib stack
--does-- support such a configuration but it must be validated, I cc
the maintainer here, in case they see something that they think is
broken under such a config.<br>
<br>
However, note that such a config is somehow problematic (broken) for
High Availability, specifically looking on IPoIB HA, say you have two
nodes, n1 and n2 connected to two subnet S1 and S2 and now the
n1/S1 link is broken and bonding does fail over to the IPoIB interface
on S2 such that n1/S2 is the active link. At this point, for n2 to
commuinicate with n1 it --must-- failover also to S2, when it would not
have to do so if S1 and S2 were the same fabric. <br>
<br>
This is only a simple (non) use case to examplify the problem here. My
take on that you better avoid rely on HA between subnets using tools
like bonding. If you use higher level HA tools it --might-- make sense
to plan for using two independent subnets.<br>
<br>
Or.<br>