The UvA DAS-2 cluster (fs2.das2.nikhef.nl, housed at Nikhef) is no
longer in production. The other four DAS-2 clusters are still available.
The next generation DAS system, DAS-3,
is now in production.
To keep the local /var/tmp/ file systems on the nodes from filling up,
at the end of the job temporary files created there by the user
will be removed by default.
In special cases where this automatic file deletion is undesirable,
variable SGE_KEEP_TMPFILES should be set to "yes" in "$HOME/.bashrc".
Sun Grid Engine has replaced PBS as the scheduler on all DAS-2 clusters.
In preparation of a full (DAS-2 wide) upgrade from
Sun Grid Engine (SGE),
half of the compute nodes of the DAS-2/VU cluster will be moved
from the PBS queues to SGE.
The user interface of SGE is quite similar to the one offered by PBS.
In addition, DAS-2's convenient "prun" command line job submission
interface has been ported to SGE as well.
To experiment with this new system, just follow the instructions
on the DAS-2 web page for
We expect that switching over to SGE will allow us to provide a
significantly more reliable job submission interface.
To report remaining issues with this new system, please contact
File server fs3.das2.ewi.tudelft.nl now has a new host certificate,
and Globus services are running again.
All DAS2 fileservers and almost all compute nodes have been upgraded.
For site-specific details, see below.
The fileservers have been upgraded to RedHat Enterprise Linux
Advanced Server, version 3 (RHEL3).
The compute nodes have been upgraded to CentOS-3, update 3.
CentOS-3.3 is a completely free version of RHEL3 running on the fileserver,
rebuilt from the same sources, and fully compatible with it.
The system should already be usable; things like PBS and Myrinet
are also working again.
It is no longer possible to login to the compute nodes using
plain password authentication. From now on this will require an ssh key.
You can either store the ssh key that you already use to login to DAS2
in your $HOME/.ssh directory, or you can create a new key there
Note that directly logging in to nodes should be the exception to the rule
anyway, since jobs will typically be started from the fileservers using
prun, qsub, or by means of Globus.
The default version of the Java build tool ant is now 1.5.2.
You can specify a different version of ant by setting
ANT_HOME in your $HOME/.antrc file.
For example, the following statement specifies version 1.6.0:
The Intel compilers do not work yet since the old license manager
does not work correctly on RHEL3. This will be fixed shortly.
The fileserver fs4 (Utrecht) and all its compute nodes have been upgraded.
The fileserver fs1 (Leiden) and almost all its compute nodes
have been upgraded.
Most previous hardware issues have been resolved as well.
The fileserver fs1 itself only runs with one cpu at the moment.
The fileserver fs2 (UvA/Nikhef) and almost all its compute nodes
have been upgraded.
The remaining problem on one node is hardware related.
The fileserver fs3 (Delft) and most of its compute nodes have been upgraded.
The remaining problems on two nodes are hardware related;
we'll try to resolve them shortly.
The system is usable again, but note that the domain of this cluster
has changed: fs3 and the compute nodes node300-node331 have moved to
domain das2.ewi.tudelft.nl (used to be das2.its.tudelft.nl).
NOTE: As a result of the domain change, the Globus services on fs3
are not accessible at the present.
A new host certificate should fix this next week.
The fileserver fs0 (VU) and most of its compute nodes have been upgraded.
The system should already be usable; things like PBS, Myrinet, Globus
are also working again.
The schedule for the installation of the new DAS2 operating system,
based on RedHat Enterprise Linux version 3, is as follows:
At the days mentioned, the file server will be brought down,
upgraded, and the previous configuration will be restored.
At that point, the file server is again accessible for simple
The compute nodes will then be reinstalled from scratch
and added to the PBS queues.
It may then still take a while before all additional services
such as Globus are functioning again.
|Fri Nov 19:||VU (fs0)
|Mon Nov 22:||VU (resolving any remaining issues)
|Tue Nov 23:||Delft (fs3)
|Wed Nov 24:||UvA/Nikhef (fs2)
|Thu Nov 25:||Leiden (fs1)
|Fri Nov 26:||Utrecht (fs4)
NOTE: we will try to leave the user data alone as much as
possible, but we might have to restore from backup in some cases.
Changes made after 18:00 the evenening before each site upgrade
could therefore be lost!
Work is underway to upgrade the operating system of DAS2
from the old RedHat 7.2 (RH72) release to
RedHat Enterprise Linux version 3 (RHEL3).
This is the most stable Linux OS release from RedHat that will be
actively supported the coming years.
Several details still have to be ironed out, but you can have
a preview of the new system by logging in to host node000.das2.cs.vu.nl,
which is acting as temporary file- and compile server.
At the moment there are four nodes in the PBS pool there:
Note that the home directories are still mounted from fs0.
Most of the binaries created under RedHat 7.2 will keep working,
with a couple of exceptions:
Myrinet binaries will have to be rebuilt due to an upgrade in
the Myrinet/GM drivers, and due to an incompatibility in the use
of the standard sytem libraries.
Some other binaries will have to be rebuilt as well.
For example, in case you get a runtime error like:
Incorrectly built binary which accesses errno or h_errno directly.
then the source code of the program was incorrectly referencing the
external "errno" without having included the proper C-header file "errno.h".
Do note that binaries constructed using the default gcc compiler
on RHEL3 will not run under the old RH72, due to their
dependency on the new system libraries.
More information about the upgrade path to the new OS release will
be added here as we gain more experience using the test system.
Within two weeks, the number of nodes in the PBS pool of node000.das2.cs.vu.nl
will be increased to contain about half of the nodes currently allocated
If our experiences are mostly positive, all of DAS2 will be upgraded
in a matter of days, but this will probably happen in November
at the earliest. This to allow sufficient time for testing
before completely switching over.
The latest stable release (3.2) of Globus has been installed on DAS-2.
At the moment it is only available on fs0, but so far it seems to run
fine. Unless last-minute showstoppers are found, it will also be
installed on fs1-fs4 tomorrow evening.
For more information about the new Globus release, see
The DAS2/VU cluster has been moved to a new room.
All nodes are running again, including Myrinet.
April 30, the DAS-2 cluster at the Vrije Universiteit
will be moved to a new location. It will therefore be unavailable
during that day, and possibly the day after.
Systems that mount filesystems from the VU DAS-2 fileserver (e.g.,
some workstations, the VU chemistry cluster, the Ibivu server,
various servers used by the Globe group) will also be unavailable
We also take the opportunity to wave DAS-1 goodbye. The few remaining
DAS-1 nodes as well as the DAS-1 fileserver (das0fs.cs.vu.nl) will be
shutdown Wednesday afternoon. The files currently stored on the
DAS-1 fileserver will be transferred to the DAS-2 for easy access
during the coming months, after which they will still be available