News for hackers. Stuff that matters.
Re:AMD's Future in Servers: New 7000-Series CPUs Launched an... by chrisseaton on Tuesday June 20, 2017 @05:33PM (#14598815)
I don't think running processes inside a container will be any more interesting for benchmarking than just running the processes normally. What overhead does a container add? A little indirection in syscalls? I would imagine the number of instructions involved in that are beyond trivial for serving a page, so I can't see how benchmarking running containers would be different than just benchmarking running your processes.

VMs - now with processor virtualisation technology I'm sure the different processor architectures do make an interesting difference there.

Re:AMD's Future in Servers: New 7000-Series CPUs Launched an... by myrandomcomment on Tuesday June 20, 2017 @11:20PM (#14600410)
In some of our scaling test where each container had an IP connecting out to a remote system we ran in to a ton of issues at scale that we did not see running the same number of processes on the bare metal. The overhead of the namespace can add up.
Re:AMD's Future in Servers: New 7000-Series CPUs Launched an... by nwmcsween on Wednesday June 21, 2017 @01:09AM (#14600816)
This is due to the what the container orchestratior uses to route traffic, check out project calico for low overhead networking.
Re:AMD's Future in Servers: New 7000-Series CPUs Launched an... by myrandomcomment on Wednesday June 21, 2017 @06:05AM (#14602417)
The issue had everything to do with how Linux handles networking when you have 1000 container interfaces on a single system. Tweaking net.ipv4.neigh.default.gc* solved most of the issues.

I am familiar with Calico and it would not have solved it.