Announcement

Collapse
No announcement yet.

Benchmarking Clear Linux With KPTI + Retpoline Support

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

  • #21
    Hi Michael,

    Thanks for the benchmark.

    I'd be interested if you could do a test of KPTI+IBRS/IBPB vs KPTI+retpoline. Also I don't understand whether retpoline is needed if you have IBRS+IBPB enabled. Is it complementary or a stopgap for Spectre variant 2?

    Also according to https://gist.github.com/woachk/2f867...f71c90cd6533e9 : "Retpoline isn't fully effective for Skylake and later microarchitectures".

    It would be great if vendors could come together and provide a proper statement, right now the situation couldn't be murkier.
    Last edited by cde1; 11 January 2018, 09:36 AM.

    Comment


    • #22
      Originally posted by rvdboom View Post
      Like it annihilates in Redis 2 Xeon Gold 6138, which are also dominated by a single Ryzen 1800X.........
      Same for Apache.
      So either the graphics should be read as the shorter the faster or these are very far from being server benches, don't they?
      From the graphs "More Is Better".

      The hardware configurations are very different regarding disks and so. Those graphs are only for comparing the performance of the same system before and after the patches. Not for comparing the performance of different systems.

      Comment


      • #23
        Originally posted by edwaleni View Post

        5% can be significant for a database that processes 24x7 or if a system processes voice traffic. As everyone says, your experience will vary.
        It continues being a 5% and far from those 30% and up to 50% initially claimed by some.

        Comment

        Working...
        X