nice discussion but disagree with the comment that E1/E3 is really going a consolidation point (too enterprise does /NOT/ like multiple specs); and as for SAS or should say U.2/U.3 multi-format going away would also be more akin to 'the king is dead, long like the king'. Basically PCIe switches and infrastructure is effectively non-existent. Yes, nvme/PCIe devices are great for small caches and to locate your data close to operational points, however in general business that is a infinitesimally small percentage of data (less than 1% from what I can see across clients in our outsourcing customer base). 99% need massive data stores of traditional data which plays into the large availability of SAS/FC to manage the 10's of thousands of drives. So interview is interesting, however it's really only for it's view into the 0.1% or less of the deployed market use case of storage. (1% of 1%).
I want PCIe-based options to take over the Enterprise... so SAS tech gets dumped and continues its price decline in the used markets. Homelabbers unite! ;)
@@MikeKirkReloaded kind of self-defeating then if that's your goal. If enterprise switches all pcie (and cheap pcie retimers, expanders,etc) then SAS will become extremely expensive and then non available as no-one will make them.
ok, so FDP looks very interesting! My first idea would be, for example, devide 2x3.2TB drives into 4 or more namespaces, and then in proxmox make zfs pool and stripe mirror these namespaces from different disks, nvme0n1+nvme1n1 etc. Would be same overall capacity as normal mirror, but will it be beneficial in terms of FDP reducing WAF etc.?
@@EdvardasSmakovas you got it right! Would stripe pools across the namespaces in each RUH and WAF should be better, especially if you are putting different velocity of data on each pool
nice discussion but disagree with the comment that E1/E3 is really going a consolidation point (too enterprise does /NOT/ like multiple specs); and as for SAS or should say U.2/U.3 multi-format going away would also be more akin to 'the king is dead, long like the king'. Basically PCIe switches and infrastructure is effectively non-existent. Yes, nvme/PCIe devices are great for small caches and to locate your data close to operational points, however in general business that is a infinitesimally small percentage of data (less than 1% from what I can see across clients in our outsourcing customer base). 99% need massive data stores of traditional data which plays into the large availability of SAS/FC to manage the 10's of thousands of drives.
So interview is interesting, however it's really only for it's view into the 0.1% or less of the deployed market use case of storage. (1% of 1%).
I want PCIe-based options to take over the Enterprise... so SAS tech gets dumped and continues its price decline in the used markets. Homelabbers unite! ;)
@@MikeKirkReloaded kind of self-defeating then if that's your goal. If enterprise switches all pcie (and cheap pcie retimers, expanders,etc) then SAS will become extremely expensive and then non available as no-one will make them.
ok, so FDP looks very interesting! My first idea would be, for example, devide 2x3.2TB drives into 4 or more namespaces, and then in proxmox make zfs pool and stripe mirror these namespaces from different disks, nvme0n1+nvme1n1 etc. Would be same overall capacity as normal mirror, but will it be beneficial in terms of FDP reducing WAF etc.?
Lots more info on FDP in the blog, should be working now 😂
@@JMHands Yes, that's where I found out more about FDP, cool stuff. So what do you you about my question? :)
@@EdvardasSmakovas you got it right! Would stripe pools across the namespaces in each RUH and WAF should be better, especially if you are putting different velocity of data on each pool
Where can FADU storage products be purchased by mere mortals?
@@locusm the WD SN861 will be easy to get in the channel. The gen 4 there are some partners like Swissbit and SMART modular DC4800