Re: Is RAC really HA on Linux

carel-jan,
good post.

there is a good chance that any unplanned outage (especially involving 
failover to stndby) will take at least 1 hour (ie by the time database 
support is called, entries in ldap  & onames servers changed; current 
active redos obtained from old primary etc). This is after the oncall dba 
has diagnosed the problem & decided we should fail o ver.

our users inform us that patient care (we are a hospital) becomes 
seriously affected once systems have been down for more then 30 mins.

even planned outages require a tremendous amount of time on our part to 
get concensus from all the users as to when is the best time. and of 
course we always have do this kind of maint around the 4:00am timeframe

in short, we are quite motivated  to providing increased up time where we 
can shutdown an instance without any discernable disruption of service

so when i talk about HA, i mean increasing the likelyhood that an outage 
of a machine (or one of our data centers) will not cause the application 
to fail.

ofcourse there is no such thing as true HA (especially with oracle in that 
rolling upgrades are not supported). which was really the question of my 
original post.

are folks finding that RAC on linux is significantly improving their 
ability to provide continuous availability of applications to end users

thanks,

steve






Carel-Jan Engel <cjpengel.dbalert@xxxxxxxxx>
09/13/2004 03:06 PM

 
        To:     evans036@xxxxxxxxxxx
        cc:     Niall Litchfield <niall.litchfield@xxxxxxxxx>, 
"oracle-l@xxxxxxxxxxxxx" 
<oracle-l@xxxxxxxxxxxxx>
        Subject:        Re: Is RAC really HA on Linux


Steve,

Using a SAN-based replication (i.e. copying all your changes on the SAN 
synchronously to a remote DR site) might cost you too much bandwidth.

Further, considering whether RAC or any other solution is HA or not is not 
an objective discussion. I'd rather take the requirements of the company 
as a starting point. What do you consider as HA? What is the Maximum time 
allowed for an unplanned outage? And what for a planned outage, and how 
many of them are allowed?

The maximum availability scenario combines RAC and Data Guard (from the 
point of view of Oracle). One of them might be enough for your situation. 
That should be investigated first, one shouldn't jump into technology as a 
panacee.

I've several customers running Data Guard alone as their HA solution. 
Compared with the SAN replication they save quite some bandwith: In stead 
of copying changed blocks (or even tracks, depending on the HW brand), 
you're copying redo information. No database blocks, nor copied archives 
need to go over your WAN/LAN, just the change vectors. You get the ablilty 
to set delays in remote redo-applying (mind the info got sent, just waits 
to get applied, giving you protection for logical errors keeping the 
zero-data loss intact) and the ability to open the standby R/O (again, 
redo keeps getting sent). When an outage of a couple of minutes is 
affordable, DG might be more helpful than RAC. It will cost you less most 
of the times.

Bottom line: figure out the requirements, investigate the disasters you'd 
like to be protected from, what donwtime is allowed with each type of 
disaster, budget, and possible solutions, before jumping into a particular 
solution that might only cover a part of the requirements and might exceed 
the budget without providing optimal protection.

Best regards,

Carel-Jan Engel

===
If you think education is expensive, try ignorance. (Derek Bok)
=== 



On Mon, 2004-09-13 at 20:28, Stephen Evans wrote: 
niall,
agreed with the single database bit. We still plan on having a stand-by 
that can provide us with zero data loss (but definitely not HA).

it is interesting that you perceive RAC as not addressing HA (but only 
scalability). and of course the db is a single point of failure in a RAC 
config (unless you mitigate that with some kind of SAN based continuous 
copy with auto failover to that too). 

so do folks generally consider (not withstanding the db as a single point 
of failure) that RAC is NOT considered high availability? I think i'm 
inclined to agree with Niall's viewpoint if we cannot do rolling upgrades 
within the cluster. From memory, oracle RAC can only withstand rolling 
upgrades if the patch is designated as such (and patchsets are NOT). 

does anyone know if future versions of oracle RAC will support rolling 
upgrades/patchsets?

hope i'm not rambling too much.

steve






Niall Litchfield <niall.litchfield@xxxxxxxxx>
09/13/2004 10:27 AM
Please respond to Niall Litchfield

 
        To:     evans036@xxxxxxxxxxx
        cc:     oracle-l@xxxxxxxxxxxxx
        Subject:        Re: Is RAC really HA on Linux


Comments inline
On Mon, 13 Sep 2004 10:11:41 -0400, Stephen Evans <evans036@xxxxxxxxxxx> 
wrote:
> i hope i addressed this right - its my first post.

looks like it! 

> i am looking at RAC to provide an HA environment on Linux (most likely
> Redhat AS3)

I think I'd view RAC as a scalability solution for Oracle rather than
an HA solution. You still only have the one database with RAC, what
happens if that DB suffers a failure?



--
http://www.freelists.org/webpage/oracle-l

Other related posts: