Pointers/Tips for Two 40-core Build Servers
We're getting some seriously nice equipment, (2) of the following boxes:
4-way 10-core Xeons (so 4x10x2 = 80 threads per box)
128GB RAM
5 x 600GB SSD
The systems will be distributed, 1000 miles apart, principally to reduce latency at each site.
Primary use case is running GNU Make.
Question (1): What's the best configuration for this hardware and network topology, to accomplish lighting fast builds? Potential benefit from distcc, ccache, given non-collocation? Best to use independently, or over network?
(2) Again, given that the systems are geographically distributed, what would be a responsible data availability schema between the two (DRBD, NFS, plain ol' rsync cron jobs?)
(3) Is there any reason I shouldn't put the SSDs at each site into RAID5s?
(4) Is there any recent (~Linux 3.x) data on multicore scalability in the kernel at this range (40core)?
Any other general tips on build servers, targeted multicore speedup, cross-site data availability?
Thanks!
|