Eli Cohen wrote:
RDMAoE allows running the IB transport protocol using Ethernet frames allowing 
the deployment of IB semantics on lossless Ethernet fabrics. RDMAoE packets are 
standard Ethernet frames with an IEEE assigned Ethertype, a GRH, unmodified IB 
transport headers and payload.
Hi Eli and the team @ Mellanox

Before going into more detailed review and comments, I'd like to try and clarify few issues.

1. GRH - is it a must per your design to have it also for unicast packets? or maybe it just simplifies things, or both? I assume you may need it for the CM logic to keep working the way it used to.

2. is there any reason not to restrict the design for supporting addr / rdma-cm based consumers? e.g in the same manner that iWARP is?

3. CM services - note that once the ULP did the address resolution, the addr / rdma-cm data structure can/has the src/dest MACs, so basically why not push all the changes to the mad code and eimplement the MADs over raw or even datagram socket?

4. regarding your rdmaoe_sa - aside from being non review-able! (since in the same patch 3/8 you move tons of code from one place to another and add changes) it seems to me an overkill. I would like to see a solution which takes advantage of the SA non existence under Ethernet, and hence route resolution becomes no-op as it is with iWARP, as for multicast join, since it is a local operation it should and can be done by the rdma-cm, no need to modify the layers below it, expect for exposing API for the rdma-cm to do so.

5. 8/8 says "Currently, each IB port has a single GID entry in its table and that 
GID entery equals the link local IPv6 address" - does your design support IPv4 as 
well, how?


To enable RDMAoE with the mlx4 driver stack, both the mlx4_en and mlx4_ib drivers must be loaded, and the netdevice for the corresponding RDMAoE port must be running.
6. it seems that your design is somehow too tightly coupled to the connectX hca and the mlx4 driver, please note that the design should allow for implementing software rdmaoe provider as well

Individual ports of a multi port HCA can be independently configured as 
Ethernet (with support for RDMAoE) or IB, as is already the case. [...] 
Following is a series of 8 patches based on version 2.6.30 of the Linux kernel
Does the mainline kernel has all the patches to do so - I wasn't sure if this is the case. can you send the instructions?
This new series reflects changes based on feedback from the community on the 
previous set of patches. The whole series is tagged v3
There's not a single mentioning of a change vs the previous versions, how do you expect someone not to treat it as v1?!

For some reason you have chosen to use cross-posting, I don't think these patches need that.

Or.


_______________________________________________
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

Reply via email to