Re: Best linux Distro 2011
On 8/11/2011 2:12 AM, Johann Spies wrote:
> * It was a HPC-cluster that I was administrating. Normal updating of
> packages was a lot slower than on Debian servers.
Cluster nodes should be booting via bootp or PXE, and NFS mounts, and
their IP addresses and hostnames assigned statically via DHCP so they're
sticky to the same node on each boot. You have a single NFS directory
containing the root filesystem which each node mounts read only. System
daemons that require writing are redirected to NFS mounts or remote
facilities (i.e. syslog). Since these are compute nodes, pretty much
all daemons are disabled anyway.
Now, with such a setup, to updates to the OS, you shut down the cluster
nodes, run the update once on the master NFS directory, then reboot the
nodes, and you're done.
Anyone booting a full blown OS from local disk or exclusive NFS share,
on compute nodes doesn't know what the heck s/he is doing, and is simply
creating him/herself a huge management mess. This is why compute
cluster specific distributions, and management tools, exist, and there
are many of them available, both free and non free. There are far too
many to list here. Google will tell you about most of them.
> * The availability of specialised packages was also a pain. Many times
> we had to build packages specially that was not available in the normal
> repositories while on Debian Servers I could only do 'apt-get
If they're specialized packages, how can you complain? By definition it
will be hit or miss as to whether any one distro will have said package.
Debian doesn't have all HPC applications, and of those it does,
many/most are likely out of date due to the 2 year release cycle. This
really depends on what kind of HPC you're doing, and which programs you use.
In summary, you argument against OpenSuSE or for Debian, WTF use as a
compute cluster node OS is flawed from the start. Neither OS is
suitable, without some serious modification and tweaking, as a compute