Re: how to multiplex ocr and voting disk

On 2006-01-20, Jonathan Lewis <jonathan_at_jlcomp.demon.co.uk> wrote:
>
> "Fabrizio Magni" <fabrizio.magni_at_mycontinent.com> wrote in message
> news:43cd12ef$0$1089$4fafbaef_at_reader1.news.tin.it...
>> Jonathan Lewis wrote:
>>>
>>> Got my own answer:
>>> it has to be an odd number of copies.
>>>
>>> So I guess no single instance can take
>>> charge unless it has exclusively acquired
>>> more than half the known copies.
>>>
>>
>> Some days ago I wished to give the same answer...
>> ...but then I changed my mind.
>>
>> I couldn't convince myself that the odd number of voting disks is enough.
>>
>> I concluded I don't know how the voting mechanism in RAC is truly working
>> (and I couldn't find documentation on the topic).
>>
>> Maybe it is the flu or I'm naturally dumb but what happens if every voting
>> disk is owned by a different node or no node owns enough disks?
>>
>> Being at home I have no chance to play with a test system so if someone
>> can post the list of file descriptors involving every voting disk grouped
>> by process (on any node)...
>>
>
> I don't know how the voting mechanism in RAC works
> either, I was just guessing. But if the requirement is to have

        Are there references to these sorts of concerns elsewhere out on the web? I am just getting my feet wet at this point and my disks are already redundant by themselves. I didn't expect that I might be getting myself into more trouble with additional redunancy.

        I'm running 10gR2 on SLES9R2 with ocfs2 as the clusterfs.

[deletia]

-- 
Negligence will never equal intent, no matter how you 
attempt to distort reality to do so. This is what separates         |||
the real butchers from average Joes (or Fritzes) caught up in      / | \
events not in their control.