Here is my Rig:
First thing: I run all calculations in Dockercontainers. So I can have multiple "Calculation Frameworks" in parallel not just the dispy stuff without interfering dependencies.
Second thing I am waiting for a new Switch!
Client/Master:
AMD EPYC 7502P
512 GB RAM
2x 1 GBit/s Ethernet (will be 2x 100 GBit/s)
Nodes:
8x
AMD EPYC 7232P (because of Thread/RAM-Channel ratio)
128 GB RAM
2x 10 GBit/s Ethernet (will be 2x 25 GBit/s)
Storage:
BeeGFS distributed over Client and all Nodes
ATM, the Problem is setting up the Nodes causes sometimes socket.timeout, but i think that it go away with the new switch.
Any questions?
My Rig
Moderator: admin