[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [Linux-cluster] SNMP support with IBM Blade Center Fence Agent


On 3/1/11, Parvez Shaikh <parvez h shaikh gmail com> wrote:
> Hi Ryan,
> What is recommended method to deal with  blade center fencing failure in
> this situation? Do I have to add another level of fencing(between blade
> center and manual) which can fence automatically(not requiring manual
> interference)?

IIRC, I had touched upon similar fencing post some time back.

AFAIK, Manual fencing is not supported by Redhat.

Having said that, manual fencing has no place in production. At best
it is ok for PHP's  salestalk POC.

The short answer: Two levels of fencing is NOT possible in blades
within the same enclosure.
Two levels of fencing is possible in two blades housed in two
different enclosures provided it does not bother other servers when
you yank the power chords from the enclosure.

Let me explain with another example.
The solution you are trying to achieve, is possible in 2 individual
physical servers in rack.

One fencing level would be the management port which I would call (for
the sake of this post) as "in-band" fencing device.

Second would be Power fencing using power strips similar to:


(Disclaimer: I get zilch from any vendor for  that matter so you can
pick the most buttery vendor you are comfortable with)

Now this I call Power fencing or "out-of-band" fencing.

Now You have the the members of clluster across racks/continents (with
a unbreakable redundant datalinks and power control links) and we can
talk about two layer fencing.

First layer or level would be the in-band fencing using
IPMI/Bladecenter management port/DRAC/ILO/ALOM/RSA etc.

Second would be from the power control network which would yank the
power chord, as it were, off the server.

So we are possibly talking about three vlans here: power control vlan,
in-band vlan and data vlan.

Hope it is clear now to you now.

And Parvez, Yes, I do happen to know couple of HA fundas -- I have
deployed and managed few RHCS clusters in the past.

But then enclosure are SPOF if the member nodes are in the same
enclosure anyway.




[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]