Which model? Makes a big difference.Any idea how Ceph might perform on the Atom C3000?
3-node PVE with, say a pair of OSD SSDs each.
Piers
Probably C3758 or C3858.Which model? Makes a big difference.
I am sketching next-gen STH hosting cluster architecture and the plan is Atom C3000 series for the Ceph cluster nodes.Probably C3758 or C3858.
The Supermicro 5019A-FTN4 and E300-9A look like they could make interesting edge converged cluster nodes.
I've thought about that too. One issue for me remains SFP+. As of now, the only SM Denverton board with SFP+ is the A2SDi-H-TP4F. The others with 10Gig are Base-T, and that requires more of an infrastructure change. This is a really nice board, but 16 cores (and the $$$) are overkill for a Ceph node.I am sketching next-gen STH hosting cluster architecture and the plan is Atom C3000 series for the Ceph cluster nodes.
I agree. The Gigabyte board we have is SFP+. The Tyan is RJ-45. Both are 16C as well.I've thought about that too. One issue for me remains SFP+. As of now, the only SM Denverton board with SFP+ is the A2SDi-H-TP4F. The others with 10Gig are Base-T, and that requires more of an infrastructure change. This is a really nice board, but 16 cores (and the $$$) are overkill for a Ceph node.
I really do wish they had at least a couple of lower core-count boards with SFP+. I guess you can always shove Mellanox cards onto the other ones, but that seems to defeat a big part of the goal with this platf0rm.
I know a pretty big organization with edge compute plans at somewhere between 1,000 and 60,000 locations (depending on how you define "edge"). This company won't deploy 10gbase-T.I agree. The Gigabyte board we have is SFP+. The Tyan is RJ-45. Both are 16C as well.
I think from a product positioning standpoint, if you see the C3000 series as an edge device, then you do base-T instead of SFP+
Okay, you got me. I think a lot of these chips and boards end up as networking or NVR boxes or similar at smaller shops.I know a pretty big organization with edge compute plans at somewhere between 1,000 and 60,000 locations (depending on how you define "edge"). This company won't deploy 10gbase-T.
I have, but still using Xeon Scalable / Xeon E5 for main compute nodes.Hey Patrick,
Have you progressed any further with your C3000 converged cluster ideas?
Piers