3 ms·
Ask HN: Diskless or Diskful
Consider an architecture where you have 10 boxes per cluster. Its enough for me to have 2 diskful boxes that distribute the os images through netboot, make other boxes log remotely to these boxes. In essence I would save cost($200 per box) by going diskless than have disks.. What are your thoughts on running a node with every box haveing its own disk versus only a few master boxes having the disk...
- dryicerx 17y agoDiskless is great if your base OS image is constantly being modified/tweaked like crazy, and you wouldn't be able to accomplish the same thing by running a shell script at boot that was pulled remotely (each node having a standard shell). The downside of going diskless is you are trading off valuable memory for the ramdisk. Memory/$ > hdd/$. Have you considered just getting cheap/small hard disks for each node, and having a central shell script that runs on all of them on boot to bring them up to task?
- dedalus 17y agoCool. The other reason I was considering this was because the network throughput is better than the disk throughput this making my I/O faster..
- CyberFonic 17y agoI have several disk-less clusters. They work very well for multiple test environments. I have a bunch of older boxen with only 100Mbit eNet and the file servers use gigE. My configuration uses bootp & PxE. For even better performance you may want to look at iSCSI, FCoE or ATA over Ethernet (AoE) I prefer the later two as they run at Layer 2 not 3 and thus are better for data center deployment. It's not just the cost saving on disks. You get reduced power consumption and less noise. You could also look at using CompactFlash cards with a CF-to-IDE adaptor, it's excellent for read-mostly system disk.
- dedalus 17y agothank you.