Announcement

Collapse
No announcement yet.

Big NUMA Servers May See Better Boot Performance With Linux 5.1 Kernel

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

  • Big NUMA Servers May See Better Boot Performance With Linux 5.1 Kernel

    Phoronix: Big NUMA Servers May See Better Boot Performance With Linux 5.1 Kernel

    A patch series queued into Linux's driver core infrastructure ahead of the 5.1 kernel cycle is set to enhance the boot performance particularly for larger NUMA servers/systems. This latest round of kernel work was another contribution to the core kernel code thanks to Intel...

    http://www.phoronix.com/scan.php?pag...ot-Improvement

  • #2
    Hmm, can I ask, what's the effin point? It's not like such servers are getting rebooted daily for starters, and to top it, certain enterprise hardware, like SAS RAID controllers could take literally several minutes sorting themselves out and starting..

    Comment


    • #3
      Originally posted by aht0 View Post
      Hmm, can I ask, what's the effin point? It's not like such servers are getting rebooted daily for starters, and to top it, certain enterprise hardware, like SAS RAID controllers could take literally several minutes sorting themselves out and starting..
      3D and video rendering farms running stuff like Arch Linux do. Workloads like that tend benefit with bleeding edge updates.

      I don't think it'll be as useful to them as it will be for system designers, OS maintainers doing bare metal work, kernel devs, people working on init systems...being able to reboot the test server faster means they might be able to go live with the changes just that much sooner. One or two reboots isn't a big deal, hundreds of reboots over weeks of testing adds up. Obviously it depends on one's workload, but people doing work that requires rebooting a lot are who benefit the most.

      Comment


      • #4
        Originally posted by aht0 View Post
        Hmm, can I ask, what's the effin point? It's not like such servers are getting rebooted daily for starters, and to top it, certain enterprise hardware, like SAS RAID controllers could take literally several minutes sorting themselves out and starting..
        Why not? It's an improvement and it doesn't hurt anything. Obviously the person who worked on it thought it was worthwhile.

        Comment


        • #5
          Guess what, even if you are used to servers starting up slow it doesn't mean it doesn't matter. Servers need regular updates like everybody else (in many scenarios, even more so!). That includes kernel updates, that includes rebooting the server on a regular basis. Now on our Dual Epyc we use kexec to shove off the "several minutes" for our SAS RAID controller. What remains? Oh right, the kernel boot…

          Comment


          • #6
            Originally posted by ypnos View Post
            Guess what, even if you are used to servers starting up slow it doesn't mean it doesn't matter. Servers need regular updates like everybody else (in many scenarios, even more so!). That includes kernel updates, that includes rebooting the server on a regular basis. Now on our Dual Epyc we use kexec to shove off the "several minutes" for our SAS RAID controller. What remains? Oh right, the kernel boot…
            Heh, not only used to servers, my own home workstation used to have a LSI hardware RAID controller and an array of SAS 15krpm drives. I usually turned the machine on, then made coffee, did some basic stretches, took a piss or whatever, and when it was ready, sat down. I could hear the SAS drives initializing even into kitchen, like F1 car accelerating and switching gears in TV, lol.. I understand though when it's part of your job and it's a constant repeat activity, it could be annoying to wait for it all the time, thanks for explanation. I just did not think there is much reason to repeatedly reboot the servers daily..

            Comment


            • #7
              Originally posted by aht0 View Post
              Hmm, can I ask, what's the effin point? It's not like such servers are getting rebooted daily for starters, and to top it, certain enterprise hardware, like SAS RAID controllers could take literally several minutes sorting themselves out and starting..
              There was an interesting talk at LCA about shaving startup time on Power8/9 hardware. While it is not something that you will run up against every day, on aggregate the time saved on reboots will pay for the engineers doing the work

              You are also more likely to apply critical security updates in a timely manner if the downtime is only a couple of minutes rather than 20!

              and those with NUMA workstations (Threadripper etc) will be more than happy for quicker boot times

              Comment


              • #8
                I don't think I have a large NUMA workstation (ASUS KCMA-D8 with 2 4386's), but if the patches will help on it at all, neat

                Comment


                • #9
                  I've got a number of high end (4S) servers with >1TB RAM running Fedora which do boot from time to get the latest kernel.
                  While it is true that BIOS + RAID + NIC startup takes ages (especially if you have 16+ 10GbE and 40GbE ports and 48 drives...) but I don't mind having a faster kernel boot.

                  That said, the NUMA locality patches are far more interesting, hopefully it'll reduce the number of processes that needs to be manually glues (via CPU affinity) to a certain node / CPU socket.

                  - Gilboa
                  DEV: Intel S2600C0, 2xE52658V2, 32GB, 4x2TB + 2x3TB, GTX1080, F28/x86_64, Dell UP3216Q 4K.
                  SRV: Intel S5520SC, 2xX5680, 36GB, 4x2TB, GTX550, F28/x86_64, Dell U2711..
                  BAK: Tyan Tempest i5400XT, 2xE5335, 8GB, 3x1.5TB, 9800GTX, F28/x86-64.
                  LAP: ASUS Strix GL502V, i7-6700HQ, 32GB, 1TB+256GB, 1070M, F29/x86_64.

                  Comment

                  Working...
                  X