Except the proposed alternative should not be cp or pv, but dd bs=4M oflag=direct,sync status=progress.
I feel like I’m taking crazy pills with all the advice in this thread, because for USB keys you will otherwise end up instantly filling the write cache… which will block the apparent progress of the copy operation (so why even use pv since all you’re doing is measuring your RAM speed and available cache size) as well as heavily slow down (even potentially partially freeze in some circumstances) the rest of your system as the kernel is running out of free pages and can’t flush caches fast enough due to the slow-ass write speeds of usb keys.
* (Alternatively there is a kernel setting somewhere to disable caching globally for a block device… but in most cases caching is good, just not when you’re flashing an ISO).
so why even use pv since all you’re doing is measuring your RAM speed and available cache size
This is probably why pv progress fills in a second but is only done after a few minutes. Nonetheless, shell redirect, cat, cp work fine and handle blocksize and cache dynamically.
Your worst case scenario never happened to me after years of using pv/cp for flashing sticks/overwriting/copying partitions, even with some …risky mount settings. Honestly doesn’t make much sense to me either. Again, dd isn’t some sort of magical safe handle to make the process progress smoothly. Like i use to say, dd is a skalpell, not a shovel.
I mean yeah, the bits end up where they should. It’s just that the speed/progress indication is near useless with pv since at the end of the copy you still need to wait for the entire write buffer to be flushed (2 GiB in my experience, which can take several minutes).
So IMO dd with at least oflag=osync,odirectis safer than cp and pv with which a newbie might forget to run sync and unplug the usb key immediately, so they’ll be missing a lot of data.
Maybe some people use dd for the wrong reason, it’s their problem, but the solution is to use dd bs=4M oflag=osync,odirect, not to use cp.
It was less useless for that purpose when cp and cat were less I/O efficient compared to dd with the appropriate block size, which isn’t as much the case now as it used to be.
Obligatory useless use of dd comment.
That was an interesting read, thank you!
I haven’t touched dd since I read that about a year ago, super interesting!
For people that use dd because they like the progress bar, I highly recommend pv.
Except the proposed alternative should not be
cporpv, butdd bs=4M oflag=direct,sync status=progress.I feel like I’m taking crazy pills with all the advice in this thread, because for USB keys you will otherwise end up instantly filling the write cache… which will block the apparent progress of the copy operation (so why even use
pvsince all you’re doing is measuring your RAM speed and available cache size) as well as heavily slow down (even potentially partially freeze in some circumstances) the rest of your system as the kernel is running out of free pages and can’t flush caches fast enough due to the slow-ass write speeds of usb keys.* (Alternatively there is a kernel setting somewhere to disable caching globally for a block device… but in most cases caching is good, just not when you’re flashing an ISO).
This is probably why pv progress fills in a second but is only done after a few minutes. Nonetheless, shell redirect, cat, cp work fine and handle blocksize and cache dynamically.
Your worst case scenario never happened to me after years of using pv/cp for flashing sticks/overwriting/copying partitions, even with some …risky mount settings. Honestly doesn’t make much sense to me either. Again, dd isn’t some sort of magical safe handle to make the process progress smoothly. Like i use to say, dd is a skalpell, not a shovel.
I mean yeah, the bits end up where they should. It’s just that the speed/progress indication is near useless with pv since at the end of the copy you still need to wait for the entire write buffer to be flushed (2 GiB in my experience, which can take several minutes).
So IMO
ddwith at leastoflag=osync,odirectis safer thancpandpvwith which a newbie might forget to runsyncand unplug the usb key immediately, so they’ll be missing a lot of data.Maybe some people use
ddfor the wrong reason, it’s their problem, but the solution is to usedd bs=4M oflag=osync,odirect, not to usecp.It was less useless for that purpose when
cpandcatwere less I/O efficient compared toddwith the appropriate block size, which isn’t as much the case now as it used to be.