Hi Yesterday evening ABF has worked fine, kernel nrjQL-desktop 3.13.1-0 was built and published into kernel 3.13 repos
http://abf-downloads.rosalinux.ru/kernels_3_13x_personal/repository/ Now it's possible to install and test kernel, for this rebuild I used the rel. number -0 If you install it, then will be possible updating to officially built for omv2014 3.13.11-1 -> to install 3.13.11-0 into OpenMandriva 2014 x86_64 # urpmi.addmedia Kernel_3.13 http://abf-downloads.rosalinux.ru/kernels_3_13x_personal/repository/openmandriva2014.0/x86_64/main/release/ # urpmi --auto-select --no-verify-rpm --auto -> to install 3.13.11-0 into OpenMandriva 2014 i586 # urpmi.addmedia Kernel_3.13 http://abf-downloads.rosalinux.ru/kernels_3_13x_personal/repository/openmandriva2014.0/i586/main/release/ # urpmi --auto-select --no-verify-rpm --auto I have installed in all my PCs where everything is working fine at 100% [nicco@localhost]$ uname -a ; uptime Linux localhost.localdomain 3.13.11-nrjQL-desktop-0omv #1 SMP PREEMPT Thu Apr 24 20:31:40 UTC 2014 x86_64 x86_64 x86_64 GNU/Linux 15:10:00 up 14:41, 0 users, load average: 0,84, 1,01, 1,13 If you install it, please, test everything and report soon your feedback thanks bye, NicCo . Il Giovedì 24 Aprile 2014 22:44, Nicolò Costanza <[email protected]> ha scritto: Hi Right now, I have received the confirm that the complete tests performed on SSD devices are positive! Few mins ago, Arianna Avanzini send me this msg: "I just want to update you on the fact that the tests with 3.13.11 on SSDs are completed successfully" I have already sent the Kernel 3.13.11 to ABF bye, NicCo - Il Giovedì 24 Aprile 2014 20:59, Nicolò Costanza <[email protected]> ha scritto: Hi You can see from the new result page: http://algo.ing.unimo.it/people/paolo/disk_sched/results.php ============ latest News ============ from freenode #openmandriva-cooker chat [18:27] NicCo2 =========== [18:27] NicCo2 Kernel 3.13.11 [18:27] NicCo2 ============================= [18:27] NicCo2 BFQ v7r3: BenchTest made with 3.13.11 by my request: below the results: [18:27] NicCo2 this msg has been sent me from Arianna Avanzini (BFQ development staff) [18:27] NicCo2 ====================================================== [18:27] NicCo2 Tests with 3.13.11 on the hard disk are completed successfully. [18:27] NicCo2 In order to have as before also results on a device of a different type, [18:27] NicCo2 we started testing SSD. This last batch should be complete by evening. [18:27] NicCo2 (in my estimation should end around 22, less than my mistakes evaluation). [18:27] NicCo2 ====================================================== Now I'm rebuilding 3.13.11 nrjQL internally with BFQv7r3: I'll install and test a bit, I'll send to ABF asap I hope this may be possible this night, max for tomorrow morning, just in time for the OMV2014 GA - --------------------------------------------------------------------- %changelog * Thu Apr 24 2014 Nicolo' Costanza <[email protected]> 3.13.11-1 + update to 3.13.11 (EOL) - stable - update patches: * tuxonice-for-linux-3.13.11-2014-04-24.patch * uksm-0.1.2.2-for-v3.13.ge.9.patch - update BFQ to v7r3 * 0001-block-cgroups-kconfig-build-bits-for-BFQ-v7r3-3.13.patch * 0002-block-introduce-the-BFQ-v7r3-I-O-sched-for-3.13.patch * 0003-block-bfq-add-Early-Queue-Merge-EQM-to-BFQ-v7r3-for-.patch - suggestion / request received by Per Øyvind Karlsen (POK) * CONFIG_ACPI_CUSTOM_DSDT=y - --------------------------------------------------------------------- P.S.: The patches for the Kernel LTS 3.10.x have been released too, it will prepared ASAP, when all the tests with 3.10 will be completed ==================================================== BFQ v7 R3 - Official Announce from Paolo Valente and Arianna Avanzini ========================================================================================= Hi, the loss of of throughput with rotational devices and random workloads, correctly pointed out by Holger Hoffstätte after the release of v7r2, got stuck in out throats :) So, as you already know, we have challenged ourselves on this problem, and v7r3 is mainly the result of this effort ... It is relatively easy to just boost the throughput as the other schedulers. In fact, it is enough to to keep the drive's internal queue full. The problem is that this causes loss of control on the order in which I/O request are served. And hence loss of control on service guarantees, such as low latency. So the actual challenge was: "how to boost the throughput and at the same time preserve service guarantees?". Very briefly, the idea we have implemented is to fill the drive's queue only when the internal scheduler of the drive would serve requests basically in the same order as BFQ. As you can see from the new result page: http://algo.ing.unimo.it/people/paolo/disk_sched/results.php thanks to this improvement, on HDDs BFQ-v7r3 now achieves the same throughput as or a higher throughput than the other schedulers also with random workloads. At the same time, BFQ-v7r3 still preserves latency guarantees. For the sake of precision, BFQ-v7r3 pays a little price in terms of service guarantees: if the internal queue of the drive is filled while all the necessary conditions hold, but at some point in time, some condition stops to hold, then it may become impossible to let requests be served in the new desired order until all the requests already queued in the device have been served. According to our experiments, the most evident practical consequence of this fact is that, with random workloads, the start-up time of some application may increase by about 0.2 seconds. The second important improvement that we introduced in v7r3 is the generalization of the mechanism used to compute the parameters of the low-latency heuristic for interactive applications, so as to fit also slower storage devices. In practice, this improvement allows the low-latency guarantees provided by BFQ to be preserved also on slower devices, such as portable hard disks, multimedia and SD cards. To release this new version of BFQ as soon as possible, we are announcing it even if Arianna, as of now, managed to prepare patches only for 3.13.0 and 3.14.0. We will publish patchsets also for other kernel versions as soon as they are ready. We will also upload the new and improved version of the benchmark suite used for the last experiments soon. I hope you will enjoy BFQ-v7r3, Paolo ========================================================================================== fist feedback sent by > Holger Hoffstätte Very very awesome! Backing off from further queueing seems to have done the trick; the comparatively low throughput I used to see esp. with btrfs (workload from hell: seeky and mixed r/w due to COW) seems gone. ========================================================================================== thanks, for Your attention bye, NicCo -
