For over two decades, Warewulf has powered HPC systems around the world. From simple “under the desk” clusters to large institutional systems at HPC centers as well as enterprises who rely on performance critical computing.
Through the evolution of Warewulf, we have seen various iterations provisioning models starting from CDROM / ISO images to Etherboot (predecessor to PXE), then PXE, and more recently iPXE, but even during these different bootloaders, Warewulf in it’s heart, has always been first and foremost a stateless provisioning system (e.g. the operating system node image is not written to any persistent storage and rather it boots from the network directly into a runtime system).
Warewulf v3 has been in production for over 6 years now as it has stabilized into a very solid and full featured solution. But over the last few years, there have been many innovations in Enterprise technologies which can (and should) be leveraged as part of Warewulf. Additionally, some of the lessons learned from Warewulf v3 architecture should be rolled into an updated architecture for provisioning management.
Leveraging this legacy of provisioning and cluster management brings us to where we are today. The next generation of Warewulf. Warewulf v4 is a complete rewrite in GoLang, taking in the legacy of what we've come to expect with Warewulf, bringing it into the present, and looking out into the future.
Warewulf v4 combines ultra scalability, flexibility, and simplicity with being light weight, non-intrusive, and a great tool for scientists and seasoned system administrators alike. Warewulf empowers you to scalably and easily manage thousands of compute resources.
One of the design tenants of Warewulf is how to scalably administrate many thousands of compute nodes. Generally speaking, operating system state introduce a surface for potential discrepancies and version creep between nodes and thus Warewulf has always gone with the "single system image" approach to clustered operating system management. This means that you can have a single management node "image". For the last 20 years, we've been calling this a Virtual Node File System (VNFS), but in hindsight, this is a container.
At its core, Warewulf v4 focuses on what has made Warewulf so widely loved: simplicity, ultra scalable, lightweight, and an easy to manage solution built for both scientists and seasons system administrators to be able to design a highly functional yet easy to maintain cluster no matter how big or small or customized it needs to be.
Warewulf uses iPXE for network boot. Typically iPXE is provided by the
operating system; but the iPXE binaries can be built with
scripts/build-ipxe.sh
. This script accepts command-line arguments that are
passed to the underlying make
process. e.g.,
echo "#!ipxe
echo Tagging with vlan 1000
vcreate --tag 1000 net0 autoboot || shell" >vlan-1000.ipxe
sh scripts/build-ipxe.sh EMBED=$(readlink -f vlan-1000.ipxe)
By default, build-ipxe.sh
will attempt to write iPXE builds to
/usr/local/share/ipxe/
. This path can be specified using the DESTDIR
environment variable. Other supported environment variables include
IPXE_BRANCH
and TARGETS
.
IPXE_BRANCH=master TARGETS=bin-arm64-efi/snponly.efi DESTDIR=. sh scripts/build-ipxe.sh
Update warewulf.conf
to use locally-built iPXE.
tftp:
enabled: true
systemd name: tftp
ipxe:
00:09: /usr/local/share/ipxe/bin-x86_64-efi-snponly.efi
00:00: /usr/local/share/ipxe/bin-x86_64-pcbios-undionly.kpxe
00:0B: /usr/local/share/ipxe/bin-arm64-efi-snponly.efi
00:07: /usr/local/share/ipxe/bin-x86_64-efi-snponly.efi