Con Kolivas [interview] released O20.2int commenting, "Not sure why I made this incorrect change in the first place, but it would hardly be noticable except in latency tests. This patch reverts it." He quickly followed with the release of O20.3int, explaining, "Yep I really hate doing this, sorry. This is what I was supposed to do in
Read on to see both of the patches, which apply to 2.6.0-test5-mm2 or an O20.1int patched kernel.
Alan Cox [interview] released 2.4.22-ac3 with the comment, "This one should be treated gently initially."
Read on for the full changelog.
The recently-released 2.6.0-test5-mm2 kernel [forum] includes a patch that add the missing Documentation/as-iosched.txt file. Currently the new file contains a note about trying out the deadline scheduler (especially when using TCQ disks or when high disk performance is required), how to select the I/O scheduler at boot time, and an explanation of the various tunables for the scheduler.
Read on for the full contents of as-iosched.txt, current as of 2.6.0-test5-mm2.
"The O1int interactivity backport is more substantial now bringing it in line with O20.1int with only the nanosecond resolution missing from the 2.6 work. CK vm hacks and swap prefetch were dropped and AA Vm addons were merged as part of the default - lack of time prevented me maintaining ck vm properly."
Read on for his full announcement email.
Marcelo Tosatti has released 2.4.23-pre4 "which contains networking updates, IA64 updates, PPC updates, USB updates, [a] bunch of knfsd fixes, [and] others." Perhaps more exciting for most stable 2.4 kernel [forum] users are the VM updates that are also in this release. Marcelo explains:
"And finally [I've merged the] most important parts of [the] -aa VM. Those changes are fixing some OOM deadlocks, [giving] better per-zone balancing and better reclaiming. The OOM killer has been removed."
These recent VM changes, including the removal of the OOM killer, were discussed earlier [story]. Marcelo points out that while performance feels much better, there is a need for many people to help out by downloading 2.4.23-pre4 and testing the recent improvements.
In late August, Alan Cox [interview] announced that he was soon to be taking a year off from kernel development to study for an MBA [story]. Concerned about the fate of Alan's -ac patchset, Bernhard Rosenkraenzer followed up by releasing the '-pac' patchset, or "ported -ac" [story].
Bero (Bernhard) has continued to release updated -pac patsets, and Alan recently took notice, commenting, "Cool stuff." Though to date Alan has continued to release his -ac patches [forum], he goes on to say, "For 23-pre-ac .. its all yours Bero."
Nick Piggin [interview] has posted v15 of his scheduler policy patches. He describes, "This was going to get high res timers, but instead fixed a bug that might be causing a few people oopses. Also very small interactivity tweaks."
Con Kolivas [interview] released what he indicates should be the last of his O(1) scheduler interactivity patches with the following comment:
"Tiny tweak to keep top two interactive levels round robin at the fastest (10ms) which keeps X smooth when another interactive task is also using bursts of cpu (eg web browser)."
The patch also includes the addition of a one line comment to kernel/sched.c noting his efforts. When asking if the comment was too much, Andrew Morton [interview] replied, "No. Thanks for all your work on this."
Alan Cox [interview] released 2.4.22-ac2 with the following comments:
"Various little fixups and tidying bits. Some of these probably want to get pushed on to Marcelo eventually - the small bits and the CMPCI update certainly."
The complete changelog follows.
Andrew Morton [interview] announced 2.6.0-test5-mm1 with only a brief comment, "Small fixes, mainly".
The full changelog follows.
Nick Piggin [interview] has release v14 of his scheduler policy patches with the following comments:
"I'm starting to attack SMP and NUMA balancing, which is silly because I only have a 2xSMP to test on (I'll try to get some NUMA time from OSDL). I have provided rollups with and without the "core policy". Everything else consists of cleanups and balancing changes, so its probably what SMP/NUMA testers will want to use to start with."