Re: High Availability Options
From: DA Morgan <damorgan_at_psoug.org>
Date: Thu, 29 May 2008 09:49:24 -0700
Message-ID: <1212079777.919034@bubbleator.drizzle.com>
>
>
> The specifications I had to work on were:
>
> Primary production environment should take < 2 minutes to recover from
> a hardware failure (RAC does this for me I believe)
> In the event we lose the primary data center, we should be able to to
> have the DR environment up in < 10 minutes. Frankly I'm not sure the
> requisite network changes would propagate in < 10 minutes in the event
> we flat out lost a data center, but those were the parameters I had to
> work with.
>
> My understanding is I can set up RAC with Oracle Standard just fine,
> so that more or less meets my production criteria (although the SAN
> does remain a single point of failure, albeit an unlikely one)
>
> The hard part though is how to make sure that my offsite DR database
> is up to data. Dataguard will do that for me, and I can live in
> asynchronous mode (in a disaster we can afford to lose a few unapplied
> logs; this is not a financial app). Dataguard though (I think)
> requires enterprise.
>
> The hard part of this spec for me at least is the offsite DR site,
> which is kind of ironic, because the other database we have in house
> is mysql and it has the opposite problem. Remote slaves are easy to
> set up (and part of the core product). There's no equivalent to RAC at
> all though (don't get me started about what mysql laughably calls
> their cluster solution).
Date: Thu, 29 May 2008 09:49:24 -0700
Message-ID: <1212079777.919034@bubbleator.drizzle.com>
Pat wrote:
> On May 28, 4:59 pm, hpuxrac <johnbhur..._at_sbcglobal.net> wrote:
>> On May 28, 2:20 pm, Pat <pat.ca..._at_service-now.com> wrote: >> >> >> >>> So, I've been asked to propose a HA solution for an Oracle database. >>> My original proposal was: >>> Production RAC cluster in data center A >>> DR Database (no cluster) in data center B >>> I originally proposed Oracle Enterprise at both sites with dataguard >>> being used to keep DR in synch with the prod cluster. >>> Naturally, that particular configuration added up to some serious >>> money (I don't recall the details, but I think we had 20+ LU's worth >>> of Oracle Enterprise at $40k a pop). Probably not surprisingly, the >>> customer came back and said "dear lord, can't you give us an HA >>> architecture for less money?" >>> Which brings me here. Are there any other best practises or >>> recommended approaches for a High Availability Oracle configuration >>> that don't rely on dataguard and Oracle Enterprise? >> You don't need EE for RAC necessarily. >> >> You don't need RAC at all necessarily for HA. Actually as Moans >> Nogood has demonstrated often people lose reliability and availability >> when implementing RAC. >> >> Start by defining exactly what HA means to the customer. How much >> downtime can they tolerate and how often?
>
>
> The specifications I had to work on were:
>
> Primary production environment should take < 2 minutes to recover from
> a hardware failure (RAC does this for me I believe)
> In the event we lose the primary data center, we should be able to to
> have the DR environment up in < 10 minutes. Frankly I'm not sure the
> requisite network changes would propagate in < 10 minutes in the event
> we flat out lost a data center, but those were the parameters I had to
> work with.
>
> My understanding is I can set up RAC with Oracle Standard just fine,
> so that more or less meets my production criteria (although the SAN
> does remain a single point of failure, albeit an unlikely one)
>
> The hard part though is how to make sure that my offsite DR database
> is up to data. Dataguard will do that for me, and I can live in
> asynchronous mode (in a disaster we can afford to lose a few unapplied
> logs; this is not a financial app). Dataguard though (I think)
> requires enterprise.
>
> The hard part of this spec for me at least is the offsite DR site,
> which is kind of ironic, because the other database we have in house
> is mysql and it has the opposite problem. Remote slaves are easy to
> set up (and part of the core product). There's no equivalent to RAC at
> all though (don't get me started about what mysql laughably calls
> their cluster solution).
Based on what you've written your only real solution is to go with the Enterprise Edition of Oracle. RAC should, unless you have Windows, failover in less than 2 seconds. Data Guard is only available with EE.
RAC is not a perfect solution for HA but is can be a very good one if your application supports node affinity ... not a lot of sharing of blocks and locks. This is something you should test before you move to implementation.
Again ... if the issue is cost ... get someone to help you negotiate the licenses. If you really need the functionality you have indicated to meet your SLA there is no better technology unless you move to zLinux.
-- Daniel A. Morgan Oracle Ace Director & Instructor University of Washington damorgan_at_x.washington.edu (replace x with u to respond) Puget Sound Oracle Users Group www.psoug.orgReceived on Thu May 29 2008 - 11:49:24 CDT