9i RAC without Raw Partitions ...

> Is this possible nowadays? I heard Veritas recently released a
> cluster file system certified for 9i RAC that does away with raw
> partitions. This sounds great. Is anyone out there using it?

IIRC, it is possible with a product such as Oracle's Cluster Filesystem.

> Another quick question -- with 9i DataGuard, how do you get the
> current unarchived log off the failed node if it's dead?

You can't get it if the primary server is down. Obviously, this means
data loss. Prior to 9i and Data Guard, you just had to live with some of
the data loss in this type of situation. Data Guard lets you configure
in one of four modes to determine how the transactions get shipped to
the Standby and what type of data loss you can put up with. Obviously, a
zero data loss configuration will take some sort of committment of extra
resources on your primary database so application performance can be
impacted negatively.

HTH,
Brian

--
===================================================================

Brian Peasland
dba@remove_spam.peasland.com

Remove the "remove_spam." from the email address to email me.


"I can give it to you cheap, quick, and good. Now pick two out of
the three"

Domenic G. wrote:

> Is this possible nowadays? I heard Veritas recently released a
> cluster file system certified for 9i RAC that does away with raw
> partitions. This sounds great. Is anyone out there using it?


Oracle Corporation themselves use it to teach the 9i RAC course. There were
bug problems when first implemented (from memory, about back in January of
this year) but those mostly seem to have been resolved, and courses since
about March have run without a hitch.

Those boxes were Sun boxes, by the way. If you're running AIX or Compaq,
other CFSes are available. And if you're running Linux or Windows, Oracle
supplies their own (free) CFS. So Veritas is not the only player in town.

>
> Another quick question -- with 9i DataGuard, how do you get the
> current unarchived log off the failed node if it's dead?


Not that quick a question, actually. Depends on how you configured Data
Guard. If you've configured any of the uniquely-9i higher level of
protection, you don't need to get the last log off the failed box, because
LGWR has been transmitting redo synchronously to the standby site. Nothing
that was generated on your old primary therefore needs to be extracted,
because it's all over on the standby already. That, of course, is the whole
point behind 9i Data Guard.

However, it is possible to configure Data Guard to run exactly like the old
7,8 and 8i standby databases, whereby ARCH is responsible for shipping redo
across from primary to standby. And in that case, you will indeed have a
current redo log which has never yet been shipped, so at the point of
failure you either (a) have to somehow get at that current redo manually or
(b) give up and activate the standby database with some missing redo (and
hence some lost transactions).

For (a), the hope is that your server room is still standing, is vaguely
habitable, and the server is at least capable of spinning up its disks.
You're then supposed to squeeze out an 'alter system archive log current',
produce one more archive, and use O/S copy commands to transfer it to the
standby.

In the event of an Earthquake, Fire, Flood, Asteroid Strike disaster here> it is however a tad unlikely that you will be able to do
this, so you just don't bother, and accept that some data will have been
lost... but at least 99% of transactions are safe over in Site B.

If 100% data protection is absolutely, positively required, then you simply
don't let the situation arise... you configure Data Guard in one of its
higher protection mode, and get LGWR to ship redo synchronously, without
having to wait for a log switch.

Potentially, that's a crippler for LGWR, but you wouldn't go that route in
the first place unless you had spent an awful lot of money on robust,
redundant, high speed network links between your two sites.

Regards
HJR
--
--------------------------------------------
See my brand new website, soon to be full of
new articles: www.dizwell.com.
Nothing much there yet, but give it time!!
--------------------------------------------

Howard,

Thanks for the detailed response. When I studied Oracle Parallel
Server in 8i, I remember them mentioning that INSERT/UPDATE/DELETE DML
failover functionality would be added in addition to SELECT. I think
I need to take both courses - DataGuard and RAC, to bring myself
up-to-date, so I can fall behind again when 10g is released.

Thanks again,

Domenic

"Howard J. Rogers" wrote in message news:<3f8f0b6f$0$28119$afc38c87@news.optusnet.com.au>...
> Domenic G. wrote:
>
> > Is this possible nowadays? I heard Veritas recently released a
> > cluster file system certified for 9i RAC that does away with raw
> > partitions. This sounds great. Is anyone out there using it?
>
>
> Oracle Corporation themselves use it to teach the 9i RAC course. There were
> bug problems when first implemented (from memory, about back in January of
> this year) but those mostly seem to have been resolved, and courses since
> about March have run without a hitch.
>
> Those boxes were Sun boxes, by the way. If you're running AIX or Compaq,
> other CFSes are available. And if you're running Linux or Windows, Oracle
> supplies their own (free) CFS. So Veritas is not the only player in town.
>
> >
> > Another quick question -- with 9i DataGuard, how do you get the
> > current unarchived log off the failed node if it's dead?
>
>
> Not that quick a question, actually. Depends on how you configured Data
> Guard. If you've configured any of the uniquely-9i higher level of
> protection, you don't need to get the last log off the failed box, because
> LGWR has been transmitting redo synchronously to the standby site. Nothing
> that was generated on your old primary therefore needs to be extracted,
> because it's all over on the standby already. That, of course, is the whole
> point behind 9i Data Guard.
>
> However, it is possible to configure Data Guard to run exactly like the old
> 7,8 and 8i standby databases, whereby ARCH is responsible for shipping redo
> across from primary to standby. And in that case, you will indeed have a
> current redo log which has never yet been shipped, so at the point of
> failure you either (a) have to somehow get at that current redo manually or
> (b) give up and activate the standby database with some missing redo (and
> hence some lost transactions).
>
> For (a), the hope is that your server room is still standing, is vaguely
> habitable, and the server is at least capable of spinning up its disks.
> You're then supposed to squeeze out an 'alter system archive log current',
> produce one more archive, and use O/S copy commands to transfer it to the
> standby.
>
> In the event of an Earthquake, Fire, Flood, Asteroid Strike > disaster here> it is however a tad unlikely that you will be able to do
> this, so you just don't bother, and accept that some data will have been
> lost... but at least 99% of transactions are safe over in Site B.
>
> If 100% data protection is absolutely, positively required, then you simply
> don't let the situation arise... you configure Data Guard in one of its
> higher protection mode, and get LGWR to ship redo synchronously, without
> having to wait for a log switch.
>
> Potentially, that's a crippler for LGWR, but you wouldn't go that route in
> the first place unless you had spent an awful lot of money on robust,
> redundant, high speed network links between your two sites.
>
> Regards
> HJR

Domenic G. wrote:

>
> Thanks for the detailed response. When I studied Oracle Parallel
> Server in 8i, I remember them mentioning that INSERT/UPDATE/DELETE DML
> failover functionality would be added in addition to SELECT. I think
> I need to take both courses - DataGuard and RAC, to bring myself
> up-to-date, so I can fall behind again when 10g is released.

Different topic. You're talking about transparent application failover, or
TAF. That's select only (all DML is lost at the point of failover), and has
been for Yonks. And is still select-only in 9iR2.

From what I've seen, it's still select-only in 10g.

Regards
HJR
--
--------------------------------------------
See my brand new website, soon to be full of
new articles: www.dizwell.com.
Nothing much there yet, but give it time!!
--------------------------------------------

"Domenic G." wrote in message
> Is this possible nowadays? I heard Veritas recently released a
> cluster file system certified for 9i RAC that does away with raw
> partitions. This sounds great. Is anyone out there using it?

Been using Veritas 3.5 DE/AC for Oracle (CFS) with RAC and are happy with
the product. Compared to my experiences with OCFS (Linux), Veritas is a lot
easier to install/configure/manage and a lot more stable. If you are
currently running an older version of VCS it might be usefull knowing that
Veritas supports in-place upgrade to the 3.5 DE/AC stack as well as
conversion of the file systems to CFS.

If your platform is Linux, one of the drawbacks with OCFS as I experience
it, is that you can't host the OH on the clustered file system. Software
maintenance/patching can be a nightmare if managing a larger number of
nodes/instances. Don't know what your timeframe is, but it might be worth
looking out for OCFS ver.2.0 which is just around the corner. Ver.2.0
supports hosting both OH and archive log destinations on the clustered file
system.

If you are willing to look at alternatives to OCFS for Linux, I've been
quite impressed with Polyserver Matrix Server. It's a more mature product
than OCFS and provides a true symmetric cluster file system.

Sten


> Another quick question -- with 9i DataGuard, how do you get the
> current unarchived log off the failed node if it's dead?
>
> Thanks in advance ...
>
> Domenic.