Announcement

Collapse
No announcement yet.

CBD Proposed For The Linux Kernel: CXL Block Device

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

  • CBD Proposed For The Linux Kernel: CXL Block Device

    Phoronix: CBD Proposed For The Linux Kernel: CXL Block Device

    The latest Compute Express Link (CXL) feature work being pursued for the mainline Linux kernel is a driver to create CXL block devices for storage. On Monday a "request for comments" patch series sent out the initial code for setting up CXL shared memory to be used as Linux block devices...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    CBD Proposed For The Linux Kernel
    Yes that kernel needs to take a smoke and relax.

    Comment


    • #3
      Originally posted by Anux View Post
      Yes that kernel needs to take a smoke and relax.
      But at least he didn't forget to post the article, so there's that

      Comment


      • #4
        I want cheap CXL RAM disks... like they used to say you could get 512GB of cheap DDR2 RAM modules for CXL awhile ago. What happened to that?

        Comment


        • #5
          I've been asking our vendors for boxes with cxl memory for years now and they don't even have cxl 1.1 boxes ready ... Disaggregated rack-level cxl stuff is then 3-5 years away in practice.
          But first cxl memory modules are now being talked about so fingers crossed we'll be able to get some bigmem boxes still in this year.

          Comment


          • #6
            Cool idea. PCI express links as network and storage infrastructure.

            Comment


            • #7
              CBD is cool, but I prefer Traditional Host Caches.

              Comment


              • #8
                Originally posted by S.Pam View Post
                Cool idea. PCI express links as network and storage infrastructure.
                This may come as a surprise to some people, but PCIe is and has always been a networking protocol stack. Same with USB, DP, and a lot of other stuff. Hell, not only is AMD's infinity fabric a mesh networking technology with routers and packets and such, but just about every on-chip "bus" used in SoCs for the last 15 years has been implemented using the same "network-on-chip" concept. Swapping out shared busses for something that looks like a bus to IP modules but is actually a switched network fabric of routers massively reduces contention and increases throughput.

                Comment

                Working...
                X