California Digital offers a variety of GNU/Linux distributions on our
Itanium2, Xeon64, and Opteron servers. We can
recommend the best distribution to meet your needs or install the
option you prefer. California Digital provides
comprehensive software support through our Total Linux
Coverage™ support program. If you opt for Red
Hat Enterprise Linux, that includes support directly
from Red Hat.
Available distibutions are as follows:
- Red Hat Enterprise Linux 4.0 ES: The ES version is available for
$800 per node on Xeon64, Xeon, Itanium2, and Opteron systems. If this
option is selected, an official copy of RHEL will be installed and the
official media, license, and documentation will be shipped with the
system. Support is hence available from Red Hat in addiiton to
- Scientific Linux 3.0: Scientific Linux can be installed free
of charge on any Xeon64 or Opteron system. Support is included from
California Digital. Scientific Linux is a good choice for users
desiring extensive compatibility with popular distributions.
- Debian GNU/Linux: Debian is available on all Xeon64,
Itanium2, and Opteron systems. California Digital's engineering team
prefers Debian, but we recognize that some users are less familiar
with Debian than with some of the other choices. Support is included
from California Digital.
Installation of RHEL or Scientific Linux will get the kernels
installed with those distributions, presently 2.4.21-24.
Debian installations receive a 2.6.9 kernel.
Large customers needing specific kernel features are invited to
make their requirements known. California Digital's kernel team can
backport specific functions to specific kernels to achieve desired
compatibility. This happens most frequently with major kernel
revisions (such as the 2.4 to 2.6 evolution) when desired features
from a newer kernel branch are incompatible with applications
requiring older kernels, libraries, or applications.
Headless Platform Management
FreeIPMI and BIOSConfig are a free software projects undertaken by
California Digital (and since joined by others) to fundamentally
change how remote systems management is achieved under GNU/Linux.
This innovative technology rests upon the libfreeipmi library that
provides low-level interaction with systems implementaing the complete
IPMI 1.5 specification. FISH, the FreeIPMI
SHell, provides a scheme-based scripting interface to
libfreeipmi and enables users to harness the power of FreeIPMI.
FreeIPMI fundamentally differs from OpenIPMI in being a user
space rather than a kernel space implementation. FreeIPMI
has been deployed on the 20 teraflop Thunder cluster at Lawrence
Livermore National Lab using California Digital 6440 and 6420 systems,
as well as on clusters of California Digital 1620 systems.
Many motherboard vendors advertise compliance but do not fully
implement the IPMI 1.5 (or 2.0) specification. For that reason,
FreeIPMI may be less powerful on other hardware.
FreeIPMI allows a user to execute commands from the GNU/Linux command
line from a remote computer to perform basic system management
functions on a remote system:
- Power on
- Power off
- Environmanetal monitoring
- System Event Log
A complete discussion of FreeIPMI is beyond the scope of this
webpage. Interested users are encouraged to visit FreeIPMI web site at
BIOSConfig allows user to save, restore, edit and replicate BIOS
settings across a cluster.
Gluster is a meta cluster distribution for HPC clusters and data
centers. Gluster converts a workstation or server distribution into a
cluster distribution. User performs a stand-alone
installation using standard off-the-shelf GNU/Linux CDs and then boots
off Gluster from an external bootable medium (like CD, USB drive..) to
start the cluster installation. Gluster installer observes the
existing installation as a bunch of disks, partitions and file system.
Cluster installation process is as simple as powering on the rest of the
nodes and letting them join the installation process. With IPMI
compliant systems, even power cycling can be automated. Post
installation prepares these nodes for clustering and additionally
allows the user to feed his own custom scripts.
Unlike OSCAR and KickStart, Gluster is distribution independent and can
re-image new or corrupted nodes from other existing nodes. In a
data center mode, Gluster can choose a master server to replicate or
optionally maintain system images on a single centralized server.