New mothership boss and ops
Things are bogging down. Issues:
- One beefy node to host VMs vs. separate physical nodes for both?
I am leaning toward the latter, but I waffle. One reason is so that ops will have direct access to its storage. It is the only "correct" way to run ZFS (to assure that writes have actually hit the disk) and so we don't have overhead in dom0 running the blkback instances. One upside of VMs is the ability to redistribute resources and make snapshots. - How many cores/RAM for each?
We have seen that both boss and ops can benefit from lots of both. boss is currently 8 cores, 32GB. ops is 8/24. I would say probably 2 x 10-16 core processors on each, 64GB+ of RAM on each. - How much disk space?
Boss is currently around 2TB, mostly in /usr/testbed. /usr/testbed is where most of the action is and we would rather be somewhere north of 4TB I would think. That is probably too large for SSDs unless we really want to spend a fortune. For best performance of spinning disks we would probably want to stripe 2-4 RAID1 pairs. For ops we are around 12TB, mostly in the ZFS pool for /users and /proj. We need to decide whether we are going to continue to emphasize the shared FS model of /users and /proj or push people more toward blockstores. For the former we would probably aim more toward 20TB+ and for the latter stick with around 12TB but get a bigger storage box! - Network connectivity?
Dual 10Gb on each would be fine, one for control net, one for filesave. Technically we could do 40Gb each todellctrl0
. - Longer warranty?
Maybe 5 years instead of 3. - Other stuff?