Given the class of Spectre and Meltdown attacks rely on someone else having the freedom to execute code on your hardware, shouldn't something like this be opt-in? There's a whole world of servers out that where Spectre is ultimately completely irrelevant in terms of a security threat, but hyperthreading is definitely not irrelevant in terms of performance.
Read reviews of hyperthreaded performance gain. It's somewhere like 0% or 10%, depending on what you're doing. Not a whole lot. Hyper threading is more like a "silicon trick gone wrong".
However, I forgot to add, for OpenBSD, it may not make that much of a difference - they've never been particularly fast, especially on SMP machines, so perhaps the impact on OpenBSD is disproportionately lower and therefore acceptable? Someone should measure this.
However, I forgot to add, for OpenBSD, it may not make that much of a difference - they've never been particularly fast, especially on SMP machines, so perhaps the impact on OpenBSD is disproportionately lower and therefore acceptable? Someone should measure this.
Measure? Measure?!!?! MEASURE???!?! Are you fucking nuts? Why would anyone want to actually measure this when we can have a 2,752-message thread based purely on random anecdotes and opinions arguing over whether there's a difference or not.
(Wanders off muttering "Measure. He wants to measure").
It's true. OpenBSD does not benefit from hyper-threading, at least on all Intel platforms I have tried. Having it off happens to be a small net-win for performance as well (a few percent on compile tests).
This isn't just true for OpenBSD or for every workload either. Your mileage obviously may vary and should be tested.
A long time ago I read that it's mostly Windows that benefits from hyper threading because it doesn't save states when it does a context switch, making switching tasks much more expensive than on Linux or *BSD.
To err is human -- to blame it on a computer is even more so.
Opt-In? (Score:4, Insightful)
Given the class of Spectre and Meltdown attacks rely on someone else having the freedom to execute code on your hardware, shouldn't something like this be opt-in? There's a whole world of servers out that where Spectre is ultimately completely irrelevant in terms of a security threat, but hyperthreading is definitely not irrelevant in terms of performance.
Re: (Score:2, Insightful)
Read reviews of hyperthreaded performance gain. It's somewhere like 0% or 10%, depending on what you're doing. Not a whole lot. Hyper threading is more like a "silicon trick gone wrong".
Re:Opt-In? (Score:2)
Re:Opt-In? (Score:5, Funny)
However, I forgot to add, for OpenBSD, it may not make that much of a difference - they've never been particularly fast, especially on SMP machines, so perhaps the impact on OpenBSD is disproportionately lower and therefore acceptable? Someone should measure this.
Measure? Measure?!!?! MEASURE???!?! Are you fucking nuts? Why would anyone want to actually measure this when we can have a 2,752-message thread based purely on random anecdotes and opinions arguing over whether there's a difference or not.
(Wanders off muttering "Measure. He wants to measure").
Re: (Score:2, Insightful)
Without data, all you have is an opinion.
I don't care about your opinion.
Re: (Score:2)
Without a sense of humor, you're a bore and a horrible person to interact with.
Re: (Score:2)
I totally agree with you there!
It's just a sign that was hanging up where I used to work.
Measure!
Re:Opt-In? (Score:5, Informative)
Re: (Score:2)
A long time ago I read that it's mostly Windows that benefits from hyper threading because it doesn't save states when it does a context switch, making switching tasks much more expensive than on Linux or *BSD.