On Tue, 06 Feb 2018 14:58:02 +0100, Peter Bortas @ Pike developers forum 10353@lyskom.lysator.liu.se wrote:
Right, then I have no general complaints left (if the USE_JEMALLOC thing Nilsson pointed out is cleared up). This looks localized enough that it shouldn't break anything.
OK, I'll modify the pull request soon after a few more compilations.
I'm curious what benchmarks have been run with this. Do you have any before and after comparisons for this Tomasz?
From cursory testing of Opera Mini transcoder where both Pike and Presto web browser engine using jemalloc on several machines for a few days I can see 15% - 20% smaller memory footprint at the cost of bigger CPU usage. But there's a lot of moving parts so these numbers should not be trusted. I'll have sound numbers in a week.
On Tue, 06 Feb 2018 15:18:16 +0100, Tomasz Jamroszczak tjamroszczak@opera.com wrote:
OK, I'll modify the pull request soon after a few more compilations.
The fixup commit is in the pull request.
Thanks. Waiting for grubba to ack it before we merge.
pike-devel@lists.lysator.liu.se