On Tuesday 08 July 2008 12:29:38 Or Gerlitz wrote: > Olaf Kirch wrote: > > As Or already noted, it would be preferable to have IB and iWARP share as > > much as possible. > > I understand there probably needs to be a different transport class for > > iWARP, but I would > > like iWARP to be some kind of "we do everything like IB except these two > > little bits" transport. > > Just to clarify, the new "fastreg as work requests" API is NOT iWARP > specific, but rather comply in 95% to both IB and iWARP specs.
Yes, that much I understand. For that reason, the rds_ib_device has two new flags: is_iwarp (local end point is iwarp device), and use_fastreg (device supports the fastreg interface, so create a mr_pool that uses fastreg instead of fmrs) > Currently, support for these verbs has been posted only for the Chelsio > driver (to be merged in 2.6.27-rc1), but Mellanox has made a comment > over the list that support for the mlx4 is planned. > > So the distinction is between devices that > > 1. don't support any sort of fastreg > 2. support the proprietary fmr verbs > 3. support the standard fastreg as work request verbs > > my understanding is that RDS does not support type one ones, and its a > question whether maintainance wise, you want to support type two ones. Indeed, choice #1 isn't supported. The patch I posted will pick fastreg whenever the device advertises it. I would have preferred to do it the other way round, because the fmr code is pretty well tested by now. Unfortunately, there doesn't seem to be a capability bit that says "I support the fmr interface", and I didn't really want to start peeking at the function pointers. > Other then that fastreg issue, and assuming the credit management issue > is solved (is it?) is there any difference between IB to iWARP which you > see as relevant to RDS v3 (p2p + rdma)? Yes. credit management is there in 1.4, and on by default. I have to say I have no clue what other differences between ib and iwarp may be relevant. I can't think of any, but that's probably just my lack of imagination. > The 5% difference are things supported by iWARP but not by IB such as > read-with-invalidate, etc Yeah. Let's get it working first, then > > _______________________________________________ > 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 > -- Olaf Kirch | --- o --- Nous sommes du soleil we love when we play [EMAIL PROTECTED] | / | \ sol.dhoop.naytheet.ah kin.ir.samse.qurax _______________________________________________ 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
