Printk Cleanups Ready For Linux 6.6 - Stepping Towards Threaded/Atomic Console Printing

[Jaxad0127]

Real Time is designed for lowest guaranteed latency.

Shouldn't that be maximum guaranteed latency? As in, you can guaranteed an interrupt will be dealt with in no more than Xms max.

"lowest guaranteed latency" sounds like "I guarantee latency will always be at least Yms."

(isn't language fun?)

[wertigon]

Shouldn't that be maximum guaranteed latency? As in, you can guaranteed an interrupt will be dealt with in no more than Xms max.

"lowest guaranteed latency" sounds like "I guarantee latency will always be at least Yms."

(isn't language fun?)

Well, lowest guaranteed latency means the lowest boundary you can do is Yms, that said RT is about keeping time (every 300 ms do this, every 200 ms is a catastrophe and so is every 400 ms) not being fast, so maximum guaranteed latency isn't exactly it, either. How about this?

Generic Computing optimizes for lowest average latency or LAL, while Real Time Computing optimizes for maximum acceptable latency or MAL

[billyswong]

Or may we say, "generic" kernel optimizes throughput, while "real time" kernel optimizes consistency and avoids fluctuation? So that, things may be slower but consistently slower.

[wertigon]

Or may we say, "generic" kernel optimizes throughput, while "real time" kernel optimizes consistency and avoids fluctuation? So that, things may be slower but consistently slower.

I think you are spot on here, actually. Throughput vs Consistency is what it's all about

[Developer12]

You clearly don't understand. When I click the email button, I want to see my emails now.

Real-time makes your workload slower, not faster. It adds overhead.

The point of real-time operation is to make things predictable, in addition to being slow.