It completely depends on the way you're going to use them.
1) You're using "
Compute and Storage Separated" scenario you'll use RDMA/IB together with SMB3 protocol on top of it with Hyper-V nodes "talking" to Scale-Out File Server setup. StarWind will run @ SoFS nodes and do [own nameless protocol]-over-iSCSI in the background to keep CSVs on SoFS nodes in synch between each other (similar to upcoming Microsoft Storage Spaces Shared Nothing but w/o "slabs", multiple re-routes and not with 4 nodes minimum, but that's other story...). StarWind does not expose iSCSI to Hyper-V with this setup.
2) You're using "
Hyper-Converged" scenario then you'll have to switch IB -> IP-over-IB (we get ±33-35 Gbps on a 56 Gb IB now in the lab, ping Bohdan for more details if you want). You're going to have offload/RDMA and so on but inside: there's no **native** support for IB or RDMA with StarWind iSCSI at this moment. From the other point it's not going to help much for a reason: talking on the same node Hyper-V -to- Hyper-V we do DMA with an own driver stack and not RDMA as TCP and iSCSI stacks are mostly bypassed and with synchronization there's own protocol over iSCSI (see above) and it uses huge blocks to move data other Etherent so RDMA latency does not really matter, raw bandwidth is fine.
bacterialbag wrote:Could you clarify this for me?
I dont understand a word what you're saying.
Im using Mellanox Connectx3-Pro's myself, and was wondering if I could use any of the offloading/RDMA/Infiniband capabilities with starwind? Or will my cards only be useable with IPoIB and all the extras go to waste?
Please help clarify what infiniband options I can and can not use.
Thanks,
- Bag