Announcement

Collapse
No announcement yet.

Glibc's Per-Thread Cache Is Helping Out Some Benchmarks

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

  • #11
    Originally posted by caligula View Post
    Maybe someone with more practical wisdom could explain what these tests do and how do they relate to real world web applications.
    redis is a real world application that is sort of a key-value store, and it's use a LOT by the microservices type of application.
    It ends up being a lot of message passing kind of load, but also memory allocations are part of that

    Comment


    • #12
      Originally posted by arjan_intel View Post
      do 32 bit systems still exist (where java and the like matter) ? Even android went 64 bit
      Any Java server application running on a new enough OS to have this feature enabled would certainly be 64-bit, yes. So address-space shouldn't be an issue, but that's not to say that increased memory use won't be a concern. Even though running a 64-bit JVM, a lot of enterprise apps run clustered over VM infrastructure, without a lot of memory available - partly because having a large number of small redundant servers is more robust, and partly because giving Java more memory to manage also gives the garbage collector more work to do.

      Comment


      • #13
        Originally posted by arjan_intel View Post

        redis is a real world application that is sort of a key-value store, and it's use a LOT by the microservices type of application.
        It ends up being a lot of message passing kind of load, but also memory allocations are part of that
        Yes I know that, but the names of these particular tests don't tell how relevant the speedup is e.g. when serving web pages. If the actual network throughput increases as much, this is a huge improvement.

        Originally posted by Delgarde
        So address-space shouldn't be an issue, but that's not to say that increased memory use won't be a concern. Even though running a 64-bit JVM, a lot of enterprise apps run clustered over VM infrastructure, without a lot of memory available
        FWIW, JVM can compress references quite well.

        Comment


        • #14
          what version of jemalloc was used ?

          Comment


          • #15
            @Michael: it would be really interesting to see how much this impacts Ryzen 1700 (not 1800x) vs i7 (not i9) so we can see how much it improves painful NUMA performance. Maybe this will help Ryzen achieve an even larger advantage with its extra threads.

            Comment


            • #16
              Originally posted by caligula View Post

              You must be joking :O Heavily threaded applications and 32-bit. Wut. In 2017. Yes, it's 2017.
              Yes, embedded systems usually use 32bit CPUs, with big memory constraints.

              Comment


              • #17
                Originally posted by newwen View Post

                Yes, embedded systems usually use 32bit CPUs, with big memory constraints.
                They also tend not to be heavily threaded

                Comment


                • #18
                  Originally posted by bug77 View Post

                  They also tend not to be heavily threaded
                  Actually these days it is more common for them to have multiple slow cores. I think it is because that is more energy efficient than a single fast core.

                  Comment


                  • #19
                    Originally posted by carewolf View Post

                    Actually these days it is more common for them to have multiple slow cores. I think it is because that is more energy efficient than a single fast core.
                    Embedded apps are threaded. Just not heavily so.

                    Comment


                    • #20
                      Originally posted by carewolf View Post

                      Actually these days it is more common for them to have multiple slow cores. I think it is because that is more energy efficient than a single fast core.
                      When you compare a Ryzen system with 64 GB of RAM with a typical ARMv7 dev board with 256 to 512 MB of RAM, how would you describe the level of threading on both systems?

                      Comment

                      Working...
                      X