14

Our company has a lot of CPU intensive operations on our servers, so the performance decrease is a concern for the organization.

We did the benchmarks, and it seems that performance is almost not affected. Initially it was stated that performance will be decreased by approx. 30% after applying the patches.

Could anyone share any data on this? Without unnecessary details like company names, etc.

Jason Holcomb
  • 188
  • 8
  • 3
    I think the claim was it could be up to 30%-ish, under specific circumstances. Not that most people would see that. – Alexander O'Mara Jan 11 '18 at 15:35
  • 3
    There are [many benchmarks](https://www.google.com/search?q=phoronix+meltdown+benchmark) out there about this topic. And it was not claimed that the performance will be decreased by approx 30% in all cases but only in specific cases (with lots of syscalls). – Steffen Ullrich Jan 11 '18 at 15:43
  • CPU intensive applications will be among the least affected. – Michael Hampton Jan 11 '18 at 21:02
  • The claim was 30% worst-case -- a worst-case that you're only likely to hit if you're running a high-performance database server. Most people doing most things will be in the "5% or less" range (for example, I saw a 3% slowdown compiling the 4.14.11 kernel on an older machine). – Mark Jan 11 '18 at 21:10

1 Answers1

13

The performance impact will vary greatly depending on what the program does.

After the Meltdown patch (KPTI) whenever there is a context switch from user-space to kernel-space there is a significant overhead because of the frequent flushing of the kernel page tables.

So this means that a program that relies on frequent context switching (like a database or an hypervisor) will see a big performance impact, while others may hardly notice any.

DevopsTux
  • 861
  • 6
  • 10