This forum is more reliable than reddit. I am building a cluster server to try out new compression algorithms that my university has found. I am going to be running Siemens NX as well as matlab and other things on this system and users will connect to it through a website and log in that way all visuals will be forwarded to the user through their browser. Bandwidth will be a huge issue. I am wondering if I PXE boot all of my compute nodes off of a master storage node and if I were to do it with LACP will the system be able to boot over a link aggregated network connection and have it still work or would it be better to go with a single 10GIG Ethernet connection to each node. Can i just cut a Ethernet switch out of the middle and just connect them directly to the host or would a switch be the best solution?
I plan on using the amd a10-7700k for the individual nodes and each of the compute nodes will have an rx-470 along with 16 gigs of ram. the storage for the devices will be stored on a NFS where the systems will boot from a PXE server in order to remove as many points of failure as possible
If i were you id get a rx 480 in each of the compute nodes they are much faster than the 470, espically if you need vram. aslo if you just going to have a dedicated gpu just use an athlon 860k. It costs alot less. unless you need the integrated GPU for somthing like console video. the best choice for the cpu's though is getting some socket 2011 xeons off of ebay and using them. you could even grab a 2011 board with integrated 10gigabit nics. somthing like the Xeon E5-2670
Well I have been given a grant of 1K from my campus to do the testing with I have a master node I just need to build compute nodes and a cheap infrastructure to have them communicate with. Also what hardware would you suggest. If I have 10 gig ethernet is that better in terms of cost and performance than using multiple gigabit ethernet connections in link aggregation