Redundant Setup to minimize downtime

Printable View

10-21-2002, 10:20 AM
ronnie

Hi,

We have OAS4.0 running as a web server and oracle 8.1.7 as a database server. This web server points to the database server.

Whenever OAS or the database server is down our website is down completely. We need to have a setup where the downtime is not much... say 1/2 hour at the max.

Whats the best way to implement redundant systems so that we have our site up and running most of the time. We are looking for redundancy for both OAS and the database server.

I suggested Standby databases for the database to the management but one of my coworkers suggested that we do Mirroring of servers... Both OAS and the database server should be mirrored on a seperate machine.(We have windows 2000)

I am not sure how good this solution is and what will be the issues we will face. I am not even sure whether its even good for the Databases( If that was the case why would oracle provide standby database feature)

Also can you please suggest the alternatives I have.

Thanks

[Edited by ronnie on 10-21-2002 at 10:22 AM]
10-21-2002, 02:55 PM
jmodic

Your coworker is correct. Go to http://www.oracle.com and search for "Fail Safe" - this is oracle's WinNT/2K sollution exactly aimed for the "cheap and efficient" high availability requiremnets you are looking for.

Basicaly you will have to setup a cluster from two of your Win2K servers (clustering software provided by Microsoft) and on top of that you'll have to instal Oracle's Fail Safe (which BTW is free of charge). Then you will add your database and application server into cluser configuration, so that in case of one server failure the other one will take controll over all its clustered resources (including database instance and/or application server).

BTW, standby database configuration is primarily ment for "disaster recovery" purposes, which is not the same as "high availability" that you are looking for.
10-21-2002, 03:39 PM
marist89

Hey, as long as budgets are unlimited, why not do both?
10-21-2002, 04:50 PM
ronnie

Thanks Guys.

Clustering basically has multiple Nodes pointing to one shared storage device where the data is
stored.

http://www.microsoft.com/windows2000...ss/cluster.asp

I might seem stupid in asking this ...
but what happens when the shared device crashes or fails. How do you make sure that this does not introduce downtime in your setup.

and also if there is one shared storage and multiple nodes will be accessing it.... wont I need to configure Oracle Parallel server ???I didnt see any mention of this in the articles on oracle's web site.

Thanks

[Edited by ronnie on 10-21-2002 at 04:56 PM]
10-21-2002, 05:08 PM
marist89

Quote:

Originally posted by ronnie
I might seem stupid in asking this ...
but what happens when the shared device crashes or fails. How do you make sure that this does not introduce downtime in your setup.

Ah, the $64,000 question. The key is to configure your disk with redundancy (ie RAID). You must look at each component and determine if it is a single point of failure. If it is, there must be a backup for it.

Quote:

and also if there is one shared storage and multiple nodes will be accessing it.... wont I need to configure Oracle Parallel server ???I didnt see any mention of this in the articles on oracle's web site.
[Edited by ronnie on 10-21-2002 at 04:56 PM]

No, failsafe (as I understand it) is a failover technology and not a parallel technology. For example, you have two boxes b0 and b1. With failsafe, b0 will handle all the requests. When b0 dies, b1 kicks in and processes requests. When b0 is alive, b1 does nothing (in terms of servicing requests).

With parallel server, both b0 and b1 will be servicing requests. When b0 goes down, b1 takes over b0's requests and continues service his own requests.
10-21-2002, 05:40 PM
jmodic

Just to add a bit on Jeff's comments and to make it more clear about what happens when one server dies in your situation.

During normal operation, your b0 will run your database and your b1 will run your application server (or vice versa). So both boxes will perform their roles just like they do now.

In case your b0 fails, the box b1 will take over its database service and it will handle both application server and the database untill you fix your box b0, when everything goes back to normal.

Similarly, in case your box b1 dies, b0 will take over its load, running both database and application server untill b1 is repaired.

This is so called active-active configuration. In this configuration it is advisable that both boxes have sufficient free resources during normal operation, so that they can take additional load of another server in case of emergency.
10-21-2002, 06:10 PM
ronnie

Thanks a lot jeff and jurij,

Can I use Oracle Failsafe with OAS 4.0.8 or its Microsoft clustering which will handle the OAS part of it.

The documentation says Oracle Fail Safe supports iAS.
Thanks
10-22-2002, 03:17 AM
jmodic

I'm realy not sure about that, but I vauguly recall that indeed only iAS can be registered with Fail Safe configuration.
10-22-2002, 06:59 AM
akhadar

OFS versions(3.1.2;3.2.1;3.3.1) on Windows 2000 for OAS supports on 11i onwards
10-23-2002, 09:56 AM
ronnie

How stable and reliable is oracle Fail Safe.

Can you please share your experiences with it.

Thanks
10-23-2002, 02:06 PM
jmodic

Excelent product IMHO.
10-23-2002, 04:29 PM
ronnie

I was just thinking about it today.

Why do we need to have oracle fail safe and Microsoft windows clusters set up in the first place if you can afford a downtime of around 15-20 minutes during normal business hours and you dont care about night time issues.

We can have 2 similar machines and a shared storage between them. One machine is switched off and you install oracle software , create the services for the database stored on the shared storage. The other idle/switched off machine also has oracle software installed and services created to point to the same database. If one fails you simply switch on the other computer.

The above setup should work if you dont need Transparent Application Failover (I am assuming here that Oracle Fail safe Provides TAF).

Please confirm if thats possible or not.

Thanks
10-23-2002, 05:30 PM
jmodic

Hey, if you look it that way, then why not simplify it a bit further?

You have two simmilar machines. One is switched off, doing nothing, actually it even doesn't have any disk installed (so you don't even have to bother with installing Oracle software and configuring services)!!!! The other one is normaly configured database server. In case it fails, you simply pull out all of its disks and pull it into the empty one, start it end everything is like nothing happens. I'd call this "manual clustering".:p

However I'm sue I would not go for this option. And for yours neither, for that matter!
10-24-2002, 08:56 AM
ronnie

Quote:

Originally posted by jmodic
Hey, if you look it that way, then why not simplify it a bit further?

You have two simmilar machines. One is switched off, doing nothing, actually it even doesn't have any disk installed (so you don't even have to bother with installing Oracle software and configuring services)!!!! The other one is normaly configured database server. In case it fails, you simply pull out all of its disks and pull it into the empty one, start it end everything is like nothing happens. I'd call this "manual clustering".:p

However I'm sue I would not go for this option. And for yours neither, for that matter!

Thanks a lot jurij,

My asking the question didnt mean that i want to go for a simple solution. It was just a thought and I was wondering whether its possible or not.

Our company does not have a lot of money and we can afford a little bit of downtime.

The solution is simple and affordable but I am just curious to know about the drawbacks of such an implementation.

I know that automatic failover is not possible in the above scenario and I would really appreciate if you list some more drawbacks.

Thanks
10-24-2002, 09:59 AM
jmodic

Speaking about affordable budget. I'm not sure if WinNT/2K cluster software has to be purchased/licensed separatly, but I know that Oracle FailSafe is free of charge. So in case Microsoft clustering is free, all you'll have to invest into cluster+FailSafe configuration is two additional network cards for interconnect! (As I understand, you allready have two servers, you allready have shared storage, so basicaly you have all HW needed)

As I understand your situation you currently have two machines (one running database, other one running OAS), both active. So if you go for your solution as I understand it (you said: "One machine is switched off and you install oracle software...."), you'll have to buy another brand new machine that will do nothing but sit and wait untill you manualy start it in the case of failure. So how can this be cheaper than simply connecting two existing machines into a cluster?
10-25-2002, 08:35 AM
gandolf989

I read somewhere that you need Win2000 advanced server to do clustering. So if you don't have it you would need to be licensed for it. If it was my site I would definitely cluster though. IMHO, even though you say you can handle the downtime. What if no one is available to fix your system. If a system is worth running than it is worth backing up and having a backup system for it. Especially if you are splitting the load accross two servers to begin with.