Announcement

Collapse
No announcement yet.

The Best Features Of The Linux 5.6 Kernel From WireGuard To Y2038 Compatibility To USB4

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #21
    Originally posted by cthart View Post

    Embedded use cases aren't niche. There are already more routers and IoT devices than general purpose computers. And if I compare the growth rates of memory in those devices, they aren't going to be anywhere near 4Gb by 2038. 32-bit is here to stay, probably forever.
    "Embedded use cases aren't niche". No shit Sherlock. In 2038 though, 32 bit will be a niche even within the embedded device market. Transistors keep getting smaller and cheaper, eventually there will be no 32bit SoC left to buy to build anything, everything will be 64bit minimum. And even those few 32bit or less cases, won't care about running modern Linux kernels by that time anyway. How many people are running 8bit or 16bit cpus even in the embedded space these days? It will be exactly like this for 32bit in 2038.

    Comment


    • #22
      Originally posted by TemplarGR View Post

      "Embedded use cases aren't niche". No shit Sherlock. In 2038 though, 32 bit will be a niche even within the embedded device market. Transistors keep getting smaller and cheaper, eventually there will be no 32bit SoC left to buy to build anything, everything will be 64bit minimum. And even those few 32bit or less cases, won't care about running modern Linux kernels by that time anyway. How many people are running 8bit or 16bit cpus even in the embedded space these days? It will be exactly like this for 32bit in 2038.
      Don't agree. 2038 is only 18 years away. We've had 32-bit CPUs since the 1960s, with 32-bit OSes only becoming mainstream in the mid-1990s. 32-bit isn't going away within 2 decades for embedded devices that don't need to address 4+GB of RAM.

      Comment


      • #23
        Originally posted by cthart View Post

        Don't agree. 2038 is only 18 years away. We've had 32-bit CPUs since the 1960s, with 32-bit OSes only becoming mainstream in the mid-1990s. 32-bit isn't going away within 2 decades for embedded devices that don't need to address 4+GB of RAM.
        Right now even 100$ phones have 4GB RAM. In 2 decades, even 10$ watches will be having 4GB of RAM.... Seriously....

        And it is not just about RAM. As the whole computer science ecosystem moves to 64bit, at some point it will be more costly to use 32bit designs while you can use budget/small 64bit designs, both in terms of silicon/manufacturing and in terms of software. 32bit or less will be reserved for legacy systems and really niche situations, and some of these don't even need to run a Linux kernel.

        Comment


        • #24
          Originally posted by cthart View Post
          32-bit isn't going away within 2 decades for embedded devices that don't need to address 4+GB of RAM.
          32 bit has nothing to do with 4gb of ram. there are many 32bit systems with more ram. and there are 64bit systems with less, because those bit widths are of virtual address space, not of physical ram. or of time_t like in subj, which has nothing to do with amount of memory
          Last edited by pal666; 29 March 2020, 12:54 PM.

          Comment


          • #25
            Originally posted by TemplarGR View Post

            "Embedded use cases aren't niche". No shit Sherlock. In 2038 though, 32 bit will be a niche even within the embedded device market. Transistors keep getting smaller and cheaper, eventually there will be no 32bit SoC left to buy to build anything, everything will be 64bit minimum. And even those few 32bit or less cases, won't care about running modern Linux kernels by that time anyway. How many people are running 8bit or 16bit cpus even in the embedded space these days? It will be exactly like this for 32bit in 2038.
            Plenty of 8 and 16 bit microcontrollers are still being used and produced. In the embedded market, in many cases MOAR performance or MOAR memory or whatever buys you nothing. For some single purpose device, it only needs to be fast enough to handle whatever task it's designed to handle. 32-bit embedded devices will still be going strong in 2038.

            Now, for general purpose computers (to which you can certainly include current smartphones), yes, 32-bit is dead.

            Comment


            • #26
              Hm, IMHO this list misses one of the coolest new features: Multipath TCP (MPTCP).

              Comment


              • #27
                Even as someone who hates 32-bit crap, we will 100% still have processors that are 32-bit being made in 2038, and we need to support them as long as there's 32-bit support in the kernel. 32-bit is flat out cheaper to manufacture, and that alone will keep it around.

                Comment


                • #28
                  Oh as one idea for benchmarks... btrfs recently got several different checksum algos. I wonder if Michael is in mood to give them a try in different configurations to see how they perform?

                  As for 32-bit systems: there're plenty of embedded applications that are fine with 32 bits - and would stay equally fine 20 years later. But it would be just plain awful if they would suddenly break due to 2038 issue. 32bit also implies smaller IC die(==cheaper IC), lower power consumption, more predictable and simple system. Which is good thing any day. Especially in embedded designs.
                  Last edited by SystemCrasher; 29 March 2020, 07:11 PM.

                  Comment


                  • #29
                    Originally posted by jabl View Post

                    Plenty of 8 and 16 bit microcontrollers are still being used and produced. In the embedded market, in many cases MOAR performance or MOAR memory or whatever buys you nothing. For some single purpose device, it only needs to be fast enough to handle whatever task it's designed to handle. 32-bit embedded devices will still be going strong in 2038.

                    Now, for general purpose computers (to which you can certainly include current smartphones), yes, 32-bit is dead.
                    Are those "plenty of 8 and 16 bit microcontrollers" using the Linux kernel? Please show me....

                    In 2038, the very few 32bit chips that will exist at that point will be just a niche. "Still going strong", you wish. And they almost certainly won't be using the Linux kernel. But anyway this discussion leads to nothing.

                    Comment


                    • #30
                      Originally posted by TemplarGR View Post

                      Are those "plenty of 8 and 16 bit microcontrollers" using the Linux kernel? Please show me....
                      That is not relevant. Your statement was, quoted:

                      How many people are running 8bit or 16bit cpus even in the embedded space these days?
                      To which the answer is, plenty.

                      In 2038, the very few 32bit chips that will exist at that point will be just a niche. "Still going strong", you wish. And they almost certainly won't be using the Linux kernel.
                      I'm sure many of those 32-bit embedded devices in 2038 will run on bare metal, many others on some simple RTOS, and yet others which require more full-featured OS services will be using Linux. And evidently Linux developers agree with me, considering they've spent a lot of time implementing 2038-safe time features for 32-bit.

                      Comment

                      Working...
                      X