KTask Revived For Providing In-Kernel Multi-Threading For CPU Intensive Tasks

[phoronix]

Phoronix: KTask Revived For Providing In-Kernel Multi-Threading For CPU Intensive Tasks

It's been just about one year since the last patch series was sent out while on Monday marked a new revision to KTask, the effort that provides a generic framework to parallelize CPU-intensive kernel work...

http://www.phoronix.com/scan.php?page=news_item&px=Linux-KTask-Kernel-Multi-Thread

[birdie]

I've read the entire news piece and I still cannot figure out why this thing is needed and why the built-in kernel scheduler cannot do it.

[GrayShade]

It sounds like a thread pool that's specialized for data parallelism, without any features like work-stealing (which may not be that useful in these scenarios). So it doesn't seem particularly fancy.

[wizard69]

I've read the entire news piece and I still cannot figure out why this thing is needed and why the built-in kernel scheduler cannot do it.

Did seem to be a bit thin.

Does anybody know know why it was dropped in the first place?

[schmidtbag]

Does anybody know know why it was dropped in the first place?

Based on the phrasing "it's great to see the code revived", I assume it was left un-maintained for a while.

[Michael]

Based on the phrasing "it's great to see the code revived", I assume it was left un-maintained for a while.

Yeah just took a year for new code revision to be published... Presumably as the Oracle dev was busy with other work.

[RavFX]

Seam nice, especially if it speed up other thing than the most useless thing to speed up (the boot, because I don't use my system for just loop reboot, in that case uefi and SCSI/SAS initialization optimization would be a bigger priority ).

[carewolf]

If it can be sped up through cpu paralization, it doesn't really belong in the kernel.

[aht0]

Seam nice, especially if it speed up other thing than the most useless thing to speed up (the boot, because I don't use my system for just loop reboot, in that case uefi and SCSI/SAS initialization optimization would be a bigger priority ).

Neither does enterprise practise boot loops. LSI's SAS RAID controllers could just themselves take like 2 minutes to initialize. It's one thing enterprise does not care about because reboots are few .