Highly Threaded Linux Software Running Under CFS Quotas See Big Performance Fix

[phoronix]

Phoronix: Highly Threaded Linux Software Running Under CFS Quotas See Big Performance Fix

Thanks to a Linux kernel fix that is likely to be back-ported to the various stable series, highly threaded software running under CFS quotas for enforcing CPU limits are about to be much faster. At least in a synthetic test case, the kernel fix yields a 30x improvement in performance...

http://www.phoronix.com/scan.php?page=news_item&px=High-Threads-CFS-Quota-Fix

[Mark Rose]

Thank you for giving this more visibility. One company estimates this bug is costing them $1.5 million, so it's not an insignificant fix.

[iker]

The number from your link is not as alarming as is sounds in the beginning, considering this company is stating really high expenses just to force upstream to act in a timely manner:

1. if it is costing them 1.5M USD, they'll have additional machines to sell/rent in a few weeks - so it's really not as expensive as it seems:

costOfBug = costOfMachine * percentageUsageBeforeFix - resellValue

maximizing resellValue: sell machines after bugfix landed
minimizing percentageUsageBeforeFix: use machine as if it was a planed expansion

2. and if it were 1.5M USD and they can't accommodate those systems, they should call RH/Canonical/etc and offer them half of that as a bounty to provide custom support while this gets upstream. This would be a classical win-win, since up-streaming would be handled fast, because the contractor wants to get rid of the maintenance burden asap.

[Volta]

Thank you for giving this more visibility. One company estimates this bug is costing them $1.5 million, so it's not an insignificant fix.

They're morons, so who cares? If the bug hit the kernel in late 2017 it proves they didn't care to do anything about it. They could stick to earlier kernel. Problem solved.

[pal666]

1. if it is costing them 1.5M USD, they'll have additional machines to sell/rent in a few weeks - so it's really not as expensive as it seems

but who will give them back sales for last two years?

[iker]

but who will give them back sales for last two years?

Nobody, since that's what my second point covers: proper risk analysis and choosing the lesser of two evils - in this case expenses. This is literally the point, which makes RH et.al. earn millions every year and moves the world forward. Not considering additional support, when you're about to pay 1.5M USD for additional hardware for a regression/bug is a form of mismanagement.

[F.Ultra]

but who will give them back sales for last two years?

Linus will probably give them a discount on the next stable kernel release.

[pal666]

Linus will probably give them a discount on the next stable kernel release.

in other words it is indeed as expensive as it seems

[Mark Rose]

Here is a graph illustrating the drop in quota throttling we observed when we deployed the fix to our production Kubernetes cluster.

[Mark Rose]

They're morons, so who cares? If the bug hit the kernel in late 2017 it proves they didn't care to do anything about it. They could stick to earlier kernel. Problem solved.

The people you're calling morons are the ones who fixed it. Running old kernels is bad advice, as they often lack security fixes.