3 ms·

I've heard that before, it doesn't seem quite credible once you factor out the performance-critical code that tries to minimise context switches, and the API us

by Arnt 9y ago

I've heard that before, it doesn't seem quite credible once you factor out the performance-critical code that tries to minimise context switches, and the API usage that involves a LAN. It seems to boil down to performance-critical untrusted code that performs many kernel calls.

dtech 9y ago
It slows down everything by default. Performance-critical is relative, an app that could serv 100 req/s might not be optimized because that is sufficient, but if clearing the cache on every context switch reduces it to 70 req/s suddenly needs expensive man-hours for optimization.
Someone1234 9y ago
> I've heard that before, it doesn't seem quite credible It doesn't seem credible that completely clearing cache and repopulating it every time a privileged context switch occurs would cause performance degradation? > It seems to boil down to performance-critical untrusted code that performs many kernel calls. You're conflating the current branch prediction hotfixes with what the person above asked about which was completely clearing cache when the context is switched into a privileged state. Those are two entirely different things. The current hotfixes try to continue to utilize many performance benefits of speculative execution, while patching out the side-channel attacks from branch prediction. The person above was asking about a complete change in the underlying model we use today, which would promote security at the cost of performance.
- Arnt 9y ago
  Would cause much performance degradation where it matters most. I run some heavy jobs today. Many CPU seconds each. I don't see that those have to suffer. They can stay married to a CPU core for seconds at a time, uninterrupted, just like today. They perform few system calls, and the ones they do perform never fail any in-kernel check. The javascript that runs in the browser suffers, sure, but that javascript has other reasons not to need several seconds of CPU. BTW, that person was me, and you misunderstand the question. Not into a privileged state. Out of. Such that an unprivileged/unstrusted process would see cache contents produced by others.

by Arnt 9y ago

dtech 9y ago

It slows down everything by default. Performance-critical is relative, an app that could serv 100 req/s might not be optimized because that is sufficient, but if clearing the cache on every context switch reduces it to 70 req/s suddenly needs expensive man-hours for optimization.

Someone1234 9y ago

> I've heard that before, it doesn't seem quite credible It doesn't seem credible that completely clearing cache and repopulating it every time a privileged context switch occurs would cause performance degradation? > It seems to boil down to performance-critical untrusted code that performs many kernel calls. You're conflating the current branch prediction hotfixes with what the person above asked about which was completely clearing cache when the context is switched into a privileged state. Those are two entirely different things. The current hotfixes try to continue to utilize many performance benefits of speculative execution, while patching out the side-channel attacks from branch prediction. The person above was asking about a complete change in the underlying model we use today, which would promote security at the cost of performance.

Arnt 9y ago
Would cause much performance degradation where it matters most. I run some heavy jobs today. Many CPU seconds each. I don't see that those have to suffer. They can stay married to a CPU core for seconds at a time, uninterrupted, just like today. They perform few system calls, and the ones they do perform never fail any in-kernel check. The javascript that runs in the browser suffers, sure, but that javascript has other reasons not to need several seconds of CPU. BTW, that person was me, and you misunderstand the question. Not into a privileged state. Out of. Such that an unprivileged/unstrusted process would see cache contents produced by others.