0
I'm building a virtualization server using Proxmox (although the Hypervisor itself is not really the question here) and have the following hardware at hand (the interesting parts): Note: The hardware is NOT available physically to me but located at a hosting company.
- 1x Xeon 2630v4 -10x2,2GHz (second socket available but not used currently)
- 256GB REG ECC RAM
- 4x SSDs 1000GB (Consumer Crucial MX300)
- 4x HDDs 4000GB (2 WD, 2 Toshiba, Enterprise-SATA)
- NO HW-Raid-Controller (I did that intentionally because of ZFS but see below).
- There are no empty bays available in the current rack, I'm stuck to the number (although I could exchange the storage theoretically).
The importance of the data/the server is not extraordinarly - means, there's no big enterprise relying on it or something like that. However, reliability is still an important point
No my thought at setting up the filesystem is the following:
- Have ZFS as a filesystem because the HW is not physically available for me (replacing the HW controller might get difficult due to this in case of failure). And to take advantage of its additional features obviously.
- RAIDz2 for the OS with the SSDs
- On the SSD pool run high-performance applications (mostly data-mining stuff most likely). These are also considered more important applications relatively to the others.
- Another RAIDz2 for the HDDs. How's the performance approximately with 4 drives there?
- Run more general VMs on the HDDs - Most likely dev/test/debug VMs that don't need that much resources/rarely used concurrently. Stuff stored there is considered less important compared to the SSD pool - Most test VMs will get reset regularly anyway (taking separate backups offsite and on FTP during config changes).
- Storage should not be an issue in the forseeable future by any means. I'm already planning in a way that I should not use more than 4TB over the timeframe of the next 5 years (for the HDD pool, for the SSDs not more than 1TB).
So, does this plan sound reasonable? I'm not sure about the HDD pool, might get a striped mirror get me considerable performance advantages there - of course at the cost of some redundancy. However as said, the data there is less important, so I might utilize the performance gain otherwise.
Open for thoughts.
Greetings Marcus
I wonder what's your use case. If you optimize for high IOPS I don't think raid is the choice for the SSD portion. Also for the HDD part, would it be archiving only or there would be some random access? Do you need SSD caching for the HDD? – user3528438 – 2017-08-25T20:10:54.810
I'm not necessarily optimizing for high IOPS (then I would certainly think about getting different SSDs first). The first priority - for the SSD part - is reliability. I probably would be fine with a mirror as the activity won't be that high in general. However, as said, the HW is not physically available to me, so I'm feeling more secure with a two-drive-redundancy. The HDD part would be differently - some backups but also some random access most likely. Don't think I'll be using caching for the HDDs and instead more looking into l2arc. – CuttingWide – 2017-08-25T20:29:33.223