Ask Your Question

Huge performance regression between kernel 4.13 and 4.14

asked 2018-01-07 09:48:00 -0500

runekl gravatar image


I recently tried to upgrade a HP DL380 G8 server with 396G memory running mysql replication and a threaded java program from F25 to F27. After upgrade, the server was not able to do what it did before at all.

The java program creates a small Lucene index and send 140 000 pre-compiled queries to if threaded. This takes at least twice as long on 4.14 as on 4.13. The kernel was 4.14.8, so this was before the PTI fix.

Fortunately, booting into the 4.13.16 from F25 (without downgrading the rest of the system) fixed the problem completely.

Any ideas about what is causing this / how to proceed to figure that out? Could some of the memory management changes in 4.14 cause the regression?

edit retag flag offensive close merge delete


Could it be meltdown? Does it happen only with 4.14.11 or with any 4.14.x kernel?

fcomida gravatar imagefcomida ( 2018-01-07 12:26:17 -0500 )edit

I just tested 4,14.8 and 4.14.6, and this was before the metdown patch was added.

runekl gravatar imagerunekl ( 2018-01-07 15:28:33 -0500 )edit

2 Answers

Sort by ยป oldest newest most voted

answered 2018-01-07 13:56:54 -0500

xtreme82 gravatar image here can be the answer to your question, good evening

edit flag offensive delete link more


Thank's for the hint. But I doubt It's meltdown since it was a problem with 4.14.8. The PTI/meltdown patch was added in 4.14.11.

runekl gravatar imagerunekl ( 2018-01-07 15:27:12 -0500 )edit

answered 2018-01-07 14:51:25 -0500

noobusinghacks gravatar image

ya IT WAS BAD. i found tainted arp cache entry's and unusual scripts running in a fedora 25 distro there were bizare performance bugs happening for the machines i found having them for the most part theve seem to have gone with this update. project zero should check out gpu's next

edit flag offensive delete link more

Question Tools



Asked: 2018-01-07 09:21:40 -0500

Seen: 202 times

Last updated: Jan 07 '18