IMHO, it was a mistake to make USB block storage use the same line of names also used for local hard disks. Sure, the block device drivers for USB mass storage internally hook into the SCSI subsystem to provide block level access, and that’s why the drives are called sd[something], but why should I as an end user have to care about that? A USB drive is very much not the same thing for me as a SCSI harddisk. A NVMe drive on the other hand, kinda sorta is, at least from a practical purpose point of view, yet NVMe drives get a completely different naming scheme.
I still made the mistake, when I sleep deprived switched if and of somehow
My then girlfriend wasn’t exactly happy, that all here photos and music, which we just moved off old CDs, that couldn’t be read correctly anymore, and I spent quite some time to finally move them
Obviously the old CDs and the backup image were thrown out/deleted just a few days earlier, because I proudly had saved the bulk of it - and being poor students having loads of storage for multiple backups wasn’t in reach.
Backing them up again to fresh CDs was on the plan, but I quickly needed a live USB stick to restore my work laptop…
Since then I’m always anxious, when working with dd. Still years later I triple check and already think through my backup restoration plan
Which is a good thing in itself, but my heart rate spikes can’t be healthy
While we’re at it, can we also rename the hard drive block devices back to hd instead of sd again? SATA might use the SCSI subsystem, but SATA ain’t SCSI.
At least sata is well on its way towards dying, so the problem will solve itself in some more years.
My machines all have nvme exclusively now, only some servers are left using sata. And I would say the type of user at risk of fucking up a dd command (which 95% of the time should be a cp command) doesn’t deal with servers. Those are also not machines you plug thumb drives into commonly.
In 5-10 years we will think of sda as the usb drive, and it’ll be a fun-fact that sda used to be the boot drive.
I have a nas with 32TB. My main pc has 2TB and my laptop 512GB. I expected to need to upgrade especially the laptop at some point, but haven’t gotten anywhere near using up that local storage without even trying.
I don’t have anything huge I couldn’t put on the nas.
At this point I could easily go 4TB on the laptop and 8TB the desktop if I needed to.
Spinning rust is comparable in speed to networking anyway, so as long as noone invents a 20TB 2.5’’ hdd that fits my laptop for otg storage, there would be no reason something would benefit from an hdd in my systems over in my nas.
Edit:
Anything affordable in ssd storage has similar prices in M.2-nvme and 2.5’'-sata format. So unless you have old hardware, I see the remaining use for sata as hdd-only.
M.2 nvme uses PCIe lanes. In the last few generations both AMD and intel were quite skimpy with their PCIe lane offering, generally their consumer CPUs have only around 20-40 lanes, with servers getting over 100.
In the default configuration, nvme gets 4 lanes, so usually your average CPU will support 5-10 M.2 nvme SSDs.
However, especially with PCIe 5.0 now common, you can get the speed of 4 PCIe 3.0 lanes in a single 5.0 lane, so you can easily split all your lanes dedicating only a single lane per SSD. In that configuration your average CPU will support 20-40 drives, with only passive adapters and splitters.
Further you can for example actively split out PCIe 5.0 lanes into 4x as many 3.0 lanes, though I have not seen that done much in practice outside of the motherboard, and certainly not cheaply. Your motherboard will however usually split out the lanes into more lower-speed lanes, especially on the lower end with only 20 lanes coming out of the CPU. In practice on even entry-level boards you should count on having over 40 lanes.
As for price, you pay about 30USD for a pcie x16 to 4 M.2 slot passive card, which brings you to 6 M.2 slots on your average motherboard.
If you run up against the slot limit, you will likely be using 4TB drives and paying at the absolute lowest a grand for the bunch. I think 30USD is an acceptable tradeoff for a 20x speedup almost everyone on this situation will be taking.
If you need more than 6 drives, where you would be looking at a pcie sata or sas card previously, you can now get x16 pcie cards that passively split out to 8 M.2 slots, though the price will likely be higher. At these scales you almost certainly go for 8TB SSDs too, bringing you to 6 grand. Looking at pricing I see a raid card for 700usd, which supports passthrough, i.e. can act as just a pcie to M.2 adapter. There are probably cheaper options, but I can’t be bother to find any.
Past that there is an announced PCIe x16 to 16 slot M.2 card, for a tad over 1000usd. That is definitely not a consumer product, hence the price for what is essentially still a glorified PCIe riser.
So if for some reason you want to add tons of drives to your (non-server) system, nvme won’t stop you.
My motherboard has 3 nvme bays.
If I saw the need, there are cheap pcie to nvme cards, since (non-sata) nvme is just directing pcie lanes to the ssd anyway.
But like I said below, I don’t even have the need to get a single ssd at the currently maximum price-effective size of 4TB, no less two or three.
In my observation putting mass storage into your pc is dying in favor of either not needing that much storage, or putting it in a nas or other internet-accessible device.
Even my non-IT friends do things like put their hdd in a usb enclosure and attach it to their (internet accessible) router.
IMHO, it was a mistake to make USB block storage use the same line of names also used for local hard disks. Sure, the block device drivers for USB mass storage internally hook into the SCSI subsystem to provide block level access, and that’s why the drives are called sd[something], but why should I as an end user have to care about that? A USB drive is very much not the same thing for me as a SCSI harddisk. A NVMe drive on the other hand, kinda sorta is, at least from a practical purpose point of view, yet NVMe drives get a completely different naming scheme.
That aside, suggest you use lsblk before dd.
Yeah lsblk, lsscsi, fdsik -l , go have a coffee, come back later and hit enter on dd
Then realize you typed the command wrong and panic when you don’t get an error.
I still made the mistake, when I sleep deprived switched if and of somehow
My then girlfriend wasn’t exactly happy, that all here photos and music, which we just moved off old CDs, that couldn’t be read correctly anymore, and I spent quite some time to finally move them
Obviously the old CDs and the backup image were thrown out/deleted just a few days earlier, because I proudly had saved the bulk of it - and being poor students having loads of storage for multiple backups wasn’t in reach.
Backing them up again to fresh CDs was on the plan, but I quickly needed a live USB stick to restore my work laptop…
Since then I’m always anxious, when working with dd. Still years later I triple check and already think through my backup restoration plan
Which is a good thing in itself, but my heart rate spikes can’t be healthy
While we’re at it, can we also rename the hard drive block devices back to
hdinstead ofsdagain? SATA might use the SCSI subsystem, but SATA ain’t SCSI.At least sata is well on its way towards dying, so the problem will solve itself in some more years.
My machines all have nvme exclusively now, only some servers are left using sata. And I would say the type of user at risk of fucking up a dd command (which 95% of the time should be a cp command) doesn’t deal with servers. Those are also not machines you plug thumb drives into commonly.
In 5-10 years we will think of sda as the usb drive, and it’ll be a fun-fact that sda used to be the boot drive.
does that mean that you dont use hard drives at all? how many storage have you got?
I have a nas with 32TB. My main pc has 2TB and my laptop 512GB. I expected to need to upgrade especially the laptop at some point, but haven’t gotten anywhere near using up that local storage without even trying.
I don’t have anything huge I couldn’t put on the nas.
At this point I could easily go 4TB on the laptop and 8TB the desktop if I needed to.
Spinning rust is comparable in speed to networking anyway, so as long as noone invents a 20TB 2.5’’ hdd that fits my laptop for otg storage, there would be no reason something would benefit from an hdd in my systems over in my nas.
Edit:
Anything affordable in ssd storage has similar prices in M.2-nvme and 2.5’'-sata format. So unless you have old hardware, I see the remaining use for sata as hdd-only.
how many M.w slots do current motherboards have? a useful property of SATA is that it’s not rare to have 6 of them
M.2 nvme uses PCIe lanes. In the last few generations both AMD and intel were quite skimpy with their PCIe lane offering, generally their consumer CPUs have only around 20-40 lanes, with servers getting over 100.
In the default configuration, nvme gets 4 lanes, so usually your average CPU will support 5-10 M.2 nvme SSDs.
However, especially with PCIe 5.0 now common, you can get the speed of 4 PCIe 3.0 lanes in a single 5.0 lane, so you can easily split all your lanes dedicating only a single lane per SSD. In that configuration your average CPU will support 20-40 drives, with only passive adapters and splitters.
Further you can for example actively split out PCIe 5.0 lanes into 4x as many 3.0 lanes, though I have not seen that done much in practice outside of the motherboard, and certainly not cheaply. Your motherboard will however usually split out the lanes into more lower-speed lanes, especially on the lower end with only 20 lanes coming out of the CPU. In practice on even entry-level boards you should count on having over 40 lanes.
As for price, you pay about 30USD for a pcie x16 to 4 M.2 slot passive card, which brings you to 6 M.2 slots on your average motherboard.
If you run up against the slot limit, you will likely be using 4TB drives and paying at the absolute lowest a grand for the bunch. I think 30USD is an acceptable tradeoff for a 20x speedup almost everyone on this situation will be taking.
If you need more than 6 drives, where you would be looking at a pcie sata or sas card previously, you can now get x16 pcie cards that passively split out to 8 M.2 slots, though the price will likely be higher. At these scales you almost certainly go for 8TB SSDs too, bringing you to 6 grand. Looking at pricing I see a raid card for 700usd, which supports passthrough, i.e. can act as just a pcie to M.2 adapter. There are probably cheaper options, but I can’t be bother to find any.
Past that there is an announced PCIe x16 to 16 slot M.2 card, for a tad over 1000usd. That is definitely not a consumer product, hence the price for what is essentially still a glorified PCIe riser.
So if for some reason you want to add tons of drives to your (non-server) system, nvme won’t stop you.
doesn’t the nas use spinning rust?
It does
S-ATA still is the only way to have more than two drives in the system.
My motherboard has 3 nvme bays.
If I saw the need, there are cheap pcie to nvme cards, since (non-sata) nvme is just directing pcie lanes to the ssd anyway.
But like I said below, I don’t even have the need to get a single ssd at the currently maximum price-effective size of 4TB, no less two or three.
In my observation putting mass storage into your pc is dying in favor of either not needing that much storage, or putting it in a nas or other internet-accessible device.
Even my non-IT friends do things like put their hdd in a usb enclosure and attach it to their (internet accessible) router.