Hacker News Reader

Comment by ashelmire | Hacker News Reader

From what I've read, this slowdown only affects syscalls, which, since they aren't usually a huge percentage of processing in the first place, should not have such an effect. You're more likely looking at a few percent at most, which is not going to be enough to make AMD outperform Intel. Let's stop the fear mongering and wait for actual metrics.

View on HN

> From what I've read, this slowdown only affects syscalls

Incorrect. It also affects interrupts and (page) faults.

Any usermode to kernel and back transition.

So this is evil for virtualization hosting, which is the major enterprise application for Intel chips.

Hosting on bare metal will become more attractive. Too bad you can't long OVH and Hetzner.

>Too bad you can't long OVH and Hetzner.

What does that even mean?

Also Hetzner just introduced some AMD Epyc server.

"Long" as a verb means to purchase their stock.

As opposed to "shorting" a stock, which means making a bet that it will go down in value.

Ah that makes sense. Thanks!

For some reason I can’t reply to ‘chrisper’ but I think ‘api’ is referring to going long in the stock market.

https://www.investopedia.com/terms/l/long.asp

> For some reason I can’t reply to ‘chrisper’

HN doesn't let you do this to new comments to avoid back-and-forth commenting that is typical in flamewars.

Coming up on 9 years here and I'm still finding out new things about how this site works. I've been wondering recently why some comments aren't replyable.

I think there is a time delay. Wait a few minutes or hours and you can reply. That cooling off period has helped me really think through my replies.

They become replyable after some amount of time. The amount of time varies based on how active the thread is and/or how deeply nested the comment is.

You can reply anyway, but you have to click on the timestamp ("X minutes ago") to do it.

Usually you can click on the <posting time> and go to a page that displays only that comment, which has a reply box even when there's no reply option on the main page.

That 100Hz timer tunable just got a lot more attractive...

Does that mean that you can get an instance on AWS and slowdown the underlying server for all others by forcing a lot of syscalls? Or how is performance distributed between tenants?

Prelim benchmarks show a significant impact (~20%) on Postgresql benchmarks.

20% when running SELECT 1; over a loopback network interface, not in real-world workloads.

The other benchmark that has generated some consternation is running 'du' on a nonstop loop.

Both of these situations are pathological cases and don't reflect real-world performance. My guess is a 5-10% performance hit on general workloads. Still significant, but nowhere near as bad as some of the numbers that are getting thrown around.

And, databases are the worst case scenario, most real-world applications are showing 1% performance impact or less.

https://www.computerbase.de/2018-01/intel-cpu-pti-sicherheit...

https://www.hardwareluxx.de/index.php/news/hardware/prozesso...

Why should we trust your guesses over numbers being thrown around?

Your last link is all gaming benchmarks, which as the article mentions are not affected much.

It's not really a worst case scenario when you consider where the majority of Intel's revenue comes from: selling their high margin server chips for use in data centers, a significant portion of which are running some kind of database.

Another quick postgres estimate [1] with lower impact and a a reply from Linus Torvalds that this values are in the range what they are expecting from the patch. "... Something around 5% performance impact of the isolation is what people are looking at. ..." [2]

[1] http://lkml.iu.edu/hypermail/linux/kernel/1801.0/01274.html [2] http://lkml.iu.edu/hypermail/linux/kernel/1801.0/01299.html

> syscalls, which, since they aren't usually a huge percentage of processing in the first place... Let's stop the fear mongering

Agreed!

(We should probably also stop overgeneralizing about the nature of computational workloads.)

Software development workflows have some of the worst syscall profiles out there. This is going to hit most of us where we live.

Reply on news.ycombinator.com