[Solved] Problems testing (hardware?) mentioned in recent video

[asquirrel]

I forget what video it was, but in a recent-ish (past month?) video from LTT they mentioned it was super hard for them to test something (storage throughput? 64 core processors? something else? I forget what, exactly). I want to say the problem was the CPU being saturated with interrupts. I know that specific problem is what they hit with their NVMe storage server, but what I am asking about was a different video and a different (or at least different angle on the same) problem.

 

I remember watching the video and thinking "Hm, how would I solve that..." because when I worked for a former employer, I actually wrote my own software to test storage performance at scale (16,000+VMs). The problem is, I can't find the video where the problem to be solved is talked about so I can do more thinking about how to solve it. Or if I even can solve it. Anyone with eidetic memory who can tell me what video to re-watch?

[Roswell]

 

I believe this is what you're referring to. The clickbait title makes it impossible to know what it's actually about.

[asquirrel]

No, that wasn't it. That's the video where he struggles with NVMe bugs. But there was a more recent video where "how the heck do we test this?" was brought up. Maybe it was in a WAN show and I'm thinking it was a normal video because I never watch the wan shows live. Hrm.

[Roswell]

11 minutes ago, asquirrel said:

No, that wasn't it. That's the video where he struggles with NVMe bugs. But there was a more recent video where "how the heck do we test this?" was brought up. Maybe it was in a WAN show and I'm thinking it was a normal video because I never watch the wan shows live. Hrm.

The 3990x review? Was it something to do with NUMA nodes on 64 core chips?

[GDRRiley]

probably this one

 

[asquirrel]

46 minutes ago, Vitamanic said:

The 3990x review? Was it something to do with NUMA nodes on 64 core chips?

Winner winner. It was around the 5 minute mark. The problem was how to test a 128t processor. And that's where I got lost in my head of 'how do you test a CPU...' because it needs to do so many different tasks. You could test the interrupt handling capability, but that *should* be tied to the CPU frequently pretty solidly (a test could quantify it though!). But really, when you say 'stress', what do you want? Do you want to see if it glitches out under high interrupts when high memory bandwidth is also requested? Are you just trying to push it to 280W TDP output and see what happens? What's the goal of the test? With storage testing it was far more straightforward. But a CPU...:ahhh: