On Tue, Mar 25, 2008 at 12:40:48PM +0000, Steffan Davies wrote: > Jarkko Hietaniemi <[email protected]> wrote at 08:06 on 2008-03-25: > > > Indeed. I've seen UNIX servers with 1+ year uptime, but sooner or later > > either a disk crash or a need to patch something urgent brings them down > > either by accident or by necessity. VMS takes uptime rather seriously. > > (I don't know for certain but I assume that one can patch a cluster one > > node at a time so that the services of the cluster remain available.) > > Same applies to a cluster of any OS, surely? (With dishonourable exceptions > for such things as Exchange in which clustering seems to consist of a set > of mechanisms by which the failure of a single machine can bring about the > demise of its peers).
If you are revering to "high availability clusters" (such as HP Service guard, SUN Cluster, or Veritas Cluster), the answer is no. The cluster itself remains up, but if a node goes down, and a service is running on such a node, the service "switches" to a different machine. Depending on the service, this may be noticeble by the client (for stateless protocols like HTTP, a typical client won't notice, but if you're halfway runing your two hour SQL query, you will notice this). Abigail
pgp0cefS86OA2.pgp
Description: PGP signature
