We operate a cluster of Sun Xeon IA32 servers with Centos 3.5. Nodes are installed via a host-specific kickstart script which is generated for each node from a master template. Thus all kickstart scripts are clones of the master, edited for hostname, IP address etc only.
The CENTOS 3.5 ISO images are exported from a server & that remote NFS mount is also configured in the master template. This works perfectly well & a fully automated installation requiring no operator intervention results.
We recently added some new Opteron nodes to give us some 64-bit capability. These of course, use a slightly different master kickstart template as they mount up a different ISO directory from the NFS server, which contains the x86_64 version of Centos 3.5.
When we install these 64-bit nodes, however, we get the message: "that directory could not be mounted from the server". On hitting OK we get the usual dialogue box which displays our chosen server IP and the correct x86_64 ISO directory, if we OK that, the NFS installation proceeds unattended as normal.
I think we saw a similar thing with redHat 9 NFS kickstart installations but it was quite variable.
Anyone seen that with CENTOS3 or RHEL3 & got any clues ?
TIA
Les
On Fri, Sep 23, 2005 at 04:22:45PM +0100, Dr R L Oswald wrote:
We operate a cluster of Sun Xeon IA32 servers with Centos 3.5. Nodes are
[...]
The CENTOS 3.5 ISO images are exported from a server & that remote NFS
[... kickstart with NFS exported x86_64 iso...]
When we install these 64-bit nodes, however, we get the message: "that directory could not be mounted from the server". On hitting OK we get the usual dialogue box which displays our chosen server IP and the correct x86_64 ISO directory, if we OK that, the NFS installation proceeds unattended as normal.
I have seen that when the NFS server was too slow to answer the fast opteron boxes.. YMMV
Tru
At 16:47 23/09/05, you wrote:
On Fri, Sep 23, 2005 at 04:22:45PM +0100, Dr R L Oswald wrote:
We operate a cluster of Sun Xeon IA32 servers with Centos 3.5. Nodes are
[...]
The CENTOS 3.5 ISO images are exported from a server & that remote NFS
[... kickstart with NFS exported x86_64 iso...]
When we install these 64-bit nodes, however, we get the message: "that directory could not be mounted from the server". On hitting OK we get the usual dialogue box which displays our chosen server IP and the correct x86_64 ISO directory, if we OK that, the NFS installation proceeds unattended as normal.
I have seen that when the NFS server was too slow to answer the fast opteron boxes.. YMMV
Hi Tru
Well NFS server is a Xeon 3GHz DP with 6GB RAM & all connections are GE into the same switch. Its is not actually NFS serving anything else but on-demand Centos installs. One would have hoped that it would cope as it does with the other 3GHz clients. I'll take a look at what limited NFS tuning can be done with the Linux NFS daemon [compared to a genuine Solaris one that is >:-} ]
However your theory could explain the variability we saw with RH9 - the NFS server is one of two front ends to a cluster, sometimes students cause a brief heavy load on the server by running compute tasks that they are not meant to & that would impact NFS server response time.
Les
Tru
Tru Huynh (CentOS-3 i386/x86_64 Package Maintenance) http://pgp.mit.edu:11371/pks/lookup?op=get&search=0xBEFA581B
CentOS mailing list CentOS@centos.org http://lists.centos.org/mailman/listinfo/centos