3 ms·
Hm. Do cloud providers need to provide up to 30% more capacity for the same price, or do customers need to purchase up to 30% more capacity to get the same spee
by Waterluvian 9y ago
Hm. Do cloud providers need to provide up to 30% more capacity for the same price, or do customers need to purchase up to 30% more capacity to get the same speed?
- BoorishBears 9y agoThis bug doesn’t magically chop all CPU performance down by almost a third.
- cjbprime 9y agoWhat does it do, then? That is approximately what the mitigation patch does, according to early averages of perf changes.
- Choco31415 9y agoPerformance is chopped 5-30%, with newer CPU's with PCID being affected less significantly. [0] BoorishBears is possibly pointing out that 30% is a worst case. [0]: https://www.phoronix.com/scan.php?page=article&item=linux-415-x86pti&num=1 https://www.phoronix.com/scan.php?page=article&item=linux-41...
- BoorishBears 9y agoThis combined with what the other comment mentions, it's a range from 5%-30% with 30% being a worst case the average user does not encounter. This is an issue, but laypeople are overblowing the effect on their everyday computing.
- BuckRogers 9y agoWhat about gaming performance? Once everything is patched[0], that is. I know no one knows, but it'll be interesting to see Intel's last remaining advantage (clock frequency) mitigated somewhat by all of this. [0]https://gist.github.com/woachk/2f86755260f2fee1baf71c90cd6533e9 https://gist.github.com/woachk/2f86755260f2fee1baf71c90cd653...
- colanderman 9y agoSyscall-heavy workloads are affected to a much greater degree than CPU-heavy workloads. (This is a direct consequence of the fact that the workaround involves a per-syscall overhead.)
- Waterluvian 9y agoYes. Which is why I'm saying "up to 30%"
- rocqua 9y agoUnless demand is fully inelastic, if you up the price fewer people will buy. As such, this should hit the cloud provider profits.
- maratd 9y ago> Unless demand is fully inelastic It's mostly inelastic. People don't buy resources unless they need to do work. Alternatives, such as on-prem hardware, are affected as well. The only work that would be affected would be the work that becomes unprofitable at a 10-30% hardware cost increase and this is probably marginal enough to be ignored.
- rocqua 9y agoI'd say most of the elasticity comes from people deciding not to migrate to the cloud at all. Rather than people deciding to stop using the cloud.
- scottlamb 9y ago> It's mostly inelastic. People don't buy resources unless they need to do work. I work on a system (at Google) with significant hardware cost. Fundamentally, the work my system is doing needs to be done. But the time we spend improving efficiency is hugely elastic. I look at a CPU profile or request flow, have an idea to improve it, look at how much machine resources we're spending, guesstimate how long it will take us to implement and maintain, and use a chart to see if my idea is worth my team's time or not.[1] If the resources get more expensive or simply impossible to acquire,[2] we'll optimize more. [1] There are other considerations (will it make the code more complex, thus increasing risk of a security/privacy flaw? what about opportunity cost given that it's hard to hire more engineers and scale a team?) but that's a reasonable mental model. [2] There's a global RAM and SSD manufacturing crunch already, and if lots of CPUs are replaced due to these vulnerabilities or simply are no longer enough, that's gonna be a big crunch as well. If you're one of the few biggest cloud providers, I think you can't just replace / add tons more hardware than planned without dramatically increasing the price per unit for everyone.
- 9y ago