Nicolas Morey-Chaisemartin a écrit : > Sasha Khapyorsky a écrit : >> On 17:03 Fri 20 Mar , Nicolas Morey-Chaisemartin wrote: >>> Good. There should be one table for each switch and each of them as one >>> entry per (port/lid) couple no? >> One minhops table per switch which consists hops per port arrays for >> *only* switches' lids (CA and router lids entries are never used). >> > > It seems to me that all the entries (wether it is CA, Switch, Router, etc..) > are used in the Ftree algorithm. > The only tests done is wether this is a reald LID or not. > > If I'm right, first test would be not to set hops if target lid is not a a > switch which would already gain a some time (and memory). > > Nicolas
Wait, what I'm saying is stupid. Fat-Tree needs the hops table to be set even for CA lics to detect if we are coming from a shorter path or not... What we don't need however is to have it set for each port the remote portgroup, but just a global switch value. Nicolas _______________________________________________ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
