XPost: comp.os.linux.advocacy, comp.os.linux.misc   
   From: ldo@nz.invalid   
      
   On Sun, 29 Sep 2024 05:27:48 -0400, Paul wrote:   
      
   > On Sun, 9/29/2024 2:28 AM, Lawrence D'Oliveiro wrote:   
   >>   
   >> On Fri, 27 Sep 2024 21:32:07 -0400, Paul wrote:   
   >>   
   >>> Once you get up to a certain size, they tend to use SSD controllers   
   >>> inside, then a USB converter connected to that. This can radically   
   >>> improve the storage characteristic. The SSD controller has static and   
   >>> dynamic wear leveling. Very few USB sticks have that in a USB   
   >>> controller (but there are some).   
   >>   
   >> The thing is, Linux has, included as standard, support for filesystems   
   >> (e.g. f2fs, ubifs) that are specifically designed for use on flash   
   >> storage, with features like wear-levelling directly built into their   
   >> storage-management algorithms.   
   >>   
   >> Consider that SSDs (including USB sticks) incorporate elaborate   
   >> interface controllers to pretend to the OS that they are disks and can   
   >> use conventional disk-centric filesystems: imagine the overhead that   
   >> would simply disappear if you could go direct to the low-level storage   
   >> and use one of these purpose-built filesystems!   
   >   
   > Nobody wants their CPU donating a couple cores, to make up for the ARM   
   > cores the SSD has. One of my SSDs has a three-core ARM, two cores are   
   > for error correction on read!   
      
   Yes, but a lot of that overhead is pretending to be a raw disk on top of   
   the log-structured flash-handling layer, which is effectively its own   
   separate low-level filesystem. Both those layers would go away if you had   
   a flash-native filesystem on top. Much less overhead overall.   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)