Re: Question for the Group

On Jun 13, 4:05 am, Michael Kraemer <M.Krae...@xxxxxx> wrote:
Ron Johnson schrieb:

On 06/12/07 01:57, Michael Kraemer wrote:
[snip]

OTOH, at our site I've seen the remains of clustered VMS systems
going down immediately as a consequence of network disturbances,
showing the "blue screen of death",

Such things happened frequently more than a decade
ago, when there was only BNC ethernet. Somebody left the cable open,
and all the VAXstations downstream died and rebooted.
An NFS-mounted Unix box just froze, but usually resumed operation when
reconnected. If the whole VMS cluster was affected, and it happened
friday afternoon, you could leave for the weekend.
Such incidents were one of the reasons to leave VMS for greener Unix
pastures.
Now we have TP cabling, and the funny thing is,
similar incidents occured several times as recent as last year.
Which raises some doubts about the claimed superiority of VMS clustering.

In my experience, the type of incident described above are simply the
result of poor cluster configuration. If you don't provide
redundency then these things will happen to anyone. The failures of
the technical staff who designed the configuration, be they local, HP/
Compaq/Digital, cannot and should not be blamed on the OS.

Since UNIX systems tend to share nothing, a "freeze" is sufficient for
them to survive, however in Real Clustering, nodes share everything
therefore a break in communication requires a much more complex
response. A proper configuration would ensure that communication
could not be interupted.

I dont know how long ago this incident occurred, but if the
VAXStations clustered over the network died, then it was probably
because they were booted over the network from the main cluster system
disk. Had they been booted from internal system disks, (and no
VOTES) they would have simply hung until the re-connection interval
timed out. If the connection was restored within this time then
they would simply have continued work.

It really pisses me off when.... (No! I won't start!)

Dave.

.

Relevant Pages

  • Re: Windows 2008 bizarre cluster issue
    ... Then we went through all the network ... having run out of troubleshooting items for the server configuration ... First I would try to break any and all teaming, ... Cluster servers are IBM blades ...
    (microsoft.public.windows.server.clustering)
  • Re: Clustering across teamed / trunked adapters
    ... Windows Server 2003 ... Network adapter teaming and server clustering ... Recommended private "Heartbeat" configuration on a cluster server ...
    (microsoft.public.windows.server.clustering)
  • Re: Cluster in IBM Bladecenter HS20
    ... > fails the whole cluster becomes unavailable. ... >> then you public network). ... >>> configuration is certified by the Windows Server Catalog. ...
    (microsoft.public.windows.server.clustering)
  • RE: Unable to Create a Server 2008 Cluster
    ... Ensure TCP/IP configuration is right on both the nodes and check if you have ... When I attempt to create the cluster it fails to create the cluster. ... for more information about network troubleshooting see ...
    (microsoft.public.windows.server.clustering)
  • Member crash after increasing max_aliasid (num
    ... We have a two node TruCluster V5.1A configuration. ... We then decide to increase the number of cluster aliases to 32. ... _crashtime: struct { ... 1114396 oracle ...
    (Tru64-UNIX-Managers)