Re: 10g RAC without vendor clusterware
- From: Koen Van Langenhove <Koen.Van_Langenhove@xxxxxxxxxxx>
- To: mhthomas <qnxodba@xxxxxxxxx>
- Date: Mon, 14 Feb 2005 11:12:06 +0100
Hi Mike,
interesting comments here, thanks.
- I have virtually no experience using ASM, so can you tell me the
outlines of such a procedure ? And could you maybe give me a few
examples of where/why you need to move between ASM and non-ASM ?
- I was aware of the asmlib issue, but as far I understand this is a
linux thing so it wasn't a big issue in our situation. But I agree that
it shows how much things have been used and tested yet. After seeing
many smaller issues during one installation and a few basic tests, my
conclusion is indeed that it isn't ready for production. ( I will list
these issues in my summary)
- I don't know why there aren't a lot of people preaching the greatness
of ASM, but if everyone is going to be just very sceptic about it
instead of testing/using it, it's not going to get any better ... You're
actually the first one that answers my questions (not only on this list)
from experience.
- Not sure what you mean with SPOF strategy for ASM, in my opinion it is
just as much a SPOF as the 'real' instances, and they are supposed not
to be SPOF's because we have more than one of it ? Like I mentioned
before, what's the difference with eg VxVM software ? I know VxVM is
probably a lot stabler for now, but that's a concern about the quality
which can relatively easily be improved, as opposed to conceptual problems.
Apart from that we're using multiple disk-arrays and ASM failure-groups
to distribute the mirrors over the arrays.
What's worse regarding ASM is that it can't store the OCR (cluster
registry), since oracle provides no cluster filesystem on solaris, we
need to use a raw device. No problem there, but there's nothing to
mirror it, apart from maybe HW RAID, but then again it's just mirrored
inside 1 box => SPOF. Solaris Volume Manager (ie disksuite) is not
supported, so because of this one alone we need to use another volume
manager, and if it's on the box, I personally prefer to use it for
everything.
Furthermore, we like to create our RMAN backups to disk, and I guess
most RMAN/RAC users know what that means when there no cluster
filesystem. Once again a happy veritas account manager.
Thanks,
Koen
mhthomas wrote:
>
>How do you like moving a db from ASM to non-ASM or vice-versa? There
>are many reasons to move one back and forth so I won't specify all the
>cases, but its not fun.
>
>How about the posting that ASMLib can not handle DW (>64K) I/O that
>was posted to the list a couple days ago? My point here is just wait
>and see how long until oracle fixes this one (in my opinion critical -
>show stopper) bug. If its long, you can tell Oracle has no priority
>(e.g. $$income) for ASM. ;-)
>
>How come (in my opionion, almost) no-one (except one popular Oracle
>employee) is consistently preaching the greatness of ASM? One guy?
>
>BTW, what is your redundant (no single point failure) strategy for ASM?
>
>I'm just curious because one of my clients has a single NetApp and
>RAC/ASM and the NetApp (and ASM) is the single point of failure until
>they buy a second NetApp. Of course, if the NetApp (or ASM) fails then
>they have data loss because they have no access to the files. I know,
>I know, not a great design, but its easy to get over-confident when it
>comes to these things. I prefer to call someone 'cautious' rather than
>'pessisimistic/unemployed/etc'.
>
>HTH
>
>Regards,
>
>Mike
>
>
>
--
Regards,
Koen
--
//www.freelists.org/webpage/oracle-l
Other related posts:
