We are organizing a Birds of a feather session [1] at this years ISC
(Wednesday, June 19th 2:45pm - 3:45pm, Room Kontrast). There will be two
presentations one by Qlustar founder Roland Fehrenbacher and a second
one by Ansgar Esztermann from the Max-Planck Institute in Göttingen. The
goal of the BoF is to bring together developers, HPC cluster admins and
hardware vendors to identify the most pressing issues to further enhance
Qlustar's suitability as an open-source full stack HPC management
solution. So if you're in Frankfurt next week, take the chance to join
us.
[1] https://2019.isc-program.com/?post_type=page&p=11&id=bof125&sess=sess237
The Qlustar release 10.1.1.4-b509f1240 is ready including a
number of security and a few bug fixes. Please
check the following web pages for details about security fixes:
https://qlustar.com/news/qsa-0510191-linux-kernel-vulnerabilitieshttps://qlustar.com/news/qsa-0510192-security-update-bundle
The following non-security related bug fixes are included:
* A kernel bug in the fuse driver could lead to the spontaneous unmounting of
filesystems in rare conditions due to false handling of error
conditions. This was fixed by the Q-Leap kernel team.
* Support for the Intel/QLogic Infinipath adapters has been readded after it
was dropped by mistake in a previous release.
The Qlustar release 10.1.1.3-b509f1235 is ready including a
number of security and bug fixes as well as a couple of enhancements. Please
check the following web pages for details about security fixes:
https://qlustar.com/news/qsa-0211191-linux-kernel-vulnerabilitieshttps://qlustar.com/news/qsa-0211192-security-update-bundle
The following non-security related enhancements/bug fixes are included:
* The Qlustar netboot process has been further improved: We now ship a dedicated
C++ based multicast-enabled image pusher 'ql-mcastd' to run on the
head-node. This fault-tolerant daemon delivers the assigned OS image
of a node to the corresponding client processes running in the further
stripped down initial RAM disk (now a mere 25MB).
There is a new config file /etc/qlustar/ql-mcastd.conf for this
daemon. It is automatically generated correctly when the DHCP config
is written.
* The nvidia image module is now based on driver version 410.104 supporting the
newest GPUs.
* Qlustar installer 10.1.1-2 was updated to run on the same kernel version as on
the finally installed system.
* QluMan has added the possibility to mass import MAC addresses for new hosts
from a file. This allows to quickly setup large numbers of compute nodes.
* QluMan fixes:
- Fix exception when opening Preferences
- Don't list initramfs as possible module
- HostTemplatesWidget: Don't allow assigning global sets to a host
template
- EnclosureView: catch KeyError when deleting from Props2Hosts
- EnclosureView: Include global template when checking host hardware
status
- CommandEditor: fix exception when moving command to group
- GPU Wizard: Fix open slurm config button
The Qlustar release 10.1.1.2-b505f1215 release is ready including a
number of security and bug fixes as well as minor enhancements. Please
check the following web pages for details about security fixes:
https://qlustar.com/news/qsa-0114191-linux-kernel-vulnerabilitieshttps://qlustar.com/news/qsa-0114192-security-update-bundle
The following non-security related changes/bug fixes are included:
* Change mechanism to disable SMT (HyperThreading): Now the only supported
method is to assign the kernel parameter 'nosmt=force' in boot
configs. The previous method proved to be unstable in certain
situations due to kernel bugs.
* Update Lustre client to version 2.12.0.
* Added lustre-2.12 image module.
The Qlustar release 10.1.1.1-b505f1206 release is ready including a
number of security and bug fixes. Please check the following web pages
for details:
https://qlustar.com/news/qsa-1212181-linux-kernel-vulnerabilitieshttps://qlustar.com/news/qsa-1212182-security-update-bundle
The following non-security related changes/bug fixes are included:
* Speed up boot process of cluster nodes
* Allow to safely disable SMT (HyperThreading)
* Update Lustre client to version 2.12 rc2. Lustre 2.11 could cause a silent
crash of lustre client machines under heavy load (Ubuntu only).
The Qlustar team is pleased to announce the immediate availability of
Qlustar 10.1.1 for download at https://qlustar.com/download. This is a
minor release within the 10.1 series. It integrates the recently
published OpenHPC [ https://openhpc.community ] 1.3.6 release into the
Qlustar CentOS platform. OpenHPC 1.3.6 ships with a gcc 8 toolchain
among other goodies.
A second major improvement is the further tuning of the Qlustar node
netboot process that cuts the size of the initial RAMdisk delivered via
tftp to a mere 35MB. This doesn't only speed up the boot process in
standard clusters, but it also makes it feasible to netboot over slow
and less reliable networks like e.g. a connection to a spaceship.
Apart from that, this release comes with a number of bug and security
fixes which were already addressed in separate security advisories as
usual at
https://www.qlustar.com/news/qsa-1129181-linux-kernel-vulnerabilitieshttps://www.qlustar.com/news/qsa-1129182-security-update-bundle
The Qlustar team is pleased to announce the immediate availability of
Qlustar 10.1 for download at https://qlustar.com/download.
Adding support for CentOS 7.5 [https://www.centos.org/] as a new fully
supported edge platform, combined with the integration of OpenHPC
[https://openhpc.community], this release marks an important milestone
in Qlustar's history. Qlustar users now have the option to run the most
popular HPC platform with the same ease-of-use they're familiar with
from our Debian/Ubuntu based platform.
The second big new feature of 10.1 is a totally revamped boot
process of Qlustar nodes: QluMan 10.1 now supports downloading the OS
image via a fast and fault-tolerant multi-cast mechanism. This can
reduce boot-time dramatically and allows for simultaneous and reliable
booting of a virtually unlimited number of cluster nodes without
increasing overall boot-time.
On top of that, Qlustar OS images are now created using squashfs with
compression. This reduces the memory footprint of the OS by roughly 66%
so that a standard compute node image with slurm and IB support consumes
a mere 160MB of RAM.
Qlustar 10.1 is a feature release with its core still based on
Ubuntu/Xenial (16.04). Highlights among the various <strong>major
component updates</strong> include Kernel 4.14.x, Slurm 17.11.9.2, CUDA
9.1 and Lustre 2.11. Please read the release notes at
https://docs.qlustar.com/en-US/Qlustar_Cluster_OS/10.1/html-single/Release_…
for more details.
The Qlustar release 10.0.0.4-b484f1151 release is ready including a
number of security and bug fixes. Please check the following web pages
for details:
https://qlustar.com/news/qsa-0717181-linux-kernel-vulnerabilitieshttps://qlustar.com/news/qsa-0717182-security-update-bundle
The following non-security related bugs in QluMan are fixed:
- Fix segfault on drag&drop in ClusterConnections dialog
- Fix exception when checking for error on slurmd replies on config updates
There is a new QluMan 10.0.0 GUI singularity image at
https://qlustar.com/download as well, including the most recent
10.0.0.26 version. The same on docker hub ...