October 16, 2008 at 4:15 am
Hi all,
I'm familar with a typical 2 node cluster (active, passive). No problem. However I have a question when it comes to a 3 node cluster, running in active, active, passive). Lets say I have 5 db's (A,B,C,D and E).
Is it a case that node A is looking after db's A,B and C and node B is looking after D and E, in effect balancing the load and allowing me to group databases into function?
Sorry if this seems like a stupid question, got myself really confused by this.
Jackal
October 16, 2008 at 7:00 am
In a cluster you are clustering the instance. So A,B,C will be in an instance on node 1 and D, E would be in an instance on node 2. Node 3 would be passive and come to life when one of the other nodes dies. Load balancing isn't done by clustering, you need other methods of doing this (network level maybe or additional software)
October 16, 2008 at 7:03 am
SQL Server Active/Active and Active/Active/Passive clustering is somewhat mislabeled.
SQL Clustering is not a load balancing solution. In an Active/Active cluster, you actually set up two named instances on both servers. One named instance is active on one server and the other named instance is active on the other server. You can also configure the passive sides of these instances onto a 3rd machine as you have suggested Active/Active/Passive.
This works because a second instance of SQL has it's own services. So, you really have two Active/Passive clusters - one of them is active on one server and the other is active on the other server.
Using a third server as the passive side of the two clusters just ensures that in the event of an active node failing that hardware recources from the other active node are not used for the failed service. If ensuring that the failure of one of the instances does not slow down the other is that important, you are better off getting two additional servers rahter than just a third server and setting up an entire second cluster.
An Active/Active/Passive setup adds complication - particularly when you have to apply service packs and fail over nodes. Keeping everything apart will be much easier and probably cheaper in the long term.
October 16, 2008 at 7:09 am
Hi thank you both for the response.
It makes sense now, it is as you put it, effectively 2 active clusters, sharing a passive box. Where I was thinking of it as being almost 1 instance.
😀 - It all makes sense now.
October 16, 2008 at 7:19 am
Hi Jackal,
have a look at following gif file it's exchange-cluster but similar way we can have sql cluster.
http://www.progent.com/images/exchange-cluster-1.gif
Also following link will help you to understand more:
http://www.eggheadcafe.com/forumarchives/SQLServerserver/Aug2005/post23633739.asp
http://sql-server-performance.com/Community/forums/t/17436.aspx
Hope this will help.
---------------------------------------------------
"Thare are only 10 types of people in the world:
Those who understand binary, and those who don't."
October 16, 2008 at 7:34 am
So assuming the same databases A,B,C,D and E.
If I could set up 3 nodes, 2 active 1 passive, then I could spread the databases over the two instances.
*although, pointed noted that it may best have 2 active and 2 passive*
I assume that I would in the above example be best splitting the data files over 2 seperate LUNS in the SAN then? otherwise I would be getting minimal disk benefit?
Thanks,
Jackal
October 16, 2008 at 7:41 am
Correct.
Think of it as two completely independant clusters. At least one of them must be a named instance (because your passive server cannot have two default instances) and they obviously have to have different instance names.
Keep everything separate if possible. Give them their own LUN's (and physical drives) on your SAN. Give them different drive letters so you can keep the installs apart on the passive cluster.
All of the separation is the first part of the maintenance nightmare you end up with for the single passive server. A second passive server would allow you to use the same drive letters and default SQL instances. Remember, you can have a single physical server with two virtual servers for the passive nodes - this can help you keep things cheap and also give you a way to reallocate resources if an active node actually fails.
October 16, 2008 at 7:47 am
Thanks for that.
It all makes sense now. Got to plenty to think about, but I guess like you said the golden rule for now is to keep things as simple as possible and think of individual clusters rather than trying to create this 'monster' cluster.
Thanks again,
Jackal.
October 16, 2008 at 9:44 am
So weird... I was looking into this on Monday and since then there have been several posts to the same subject.
Does see strange the ways it’s labelled. You would assume a load-balancing cluster but infact it uses two active/passive clusters sharing the one failover node.
So your failover node becomes a member of two clusters (could get tricky for correct SAN volume addressing) that’s why we have decided to just go with two active/passive clusters of two boxes each.
Tins cheap and you don’t need to license the passive anyway.
🙂
Adam Zacks-------------------------------------------Be Nice, Or Leave
October 16, 2008 at 9:48 am
E'oh 🙁
Just seen everyones response, that'll teach me for not hitting F5.
Sorry guys, didn't mean to go repeating everyone else's point.
Let me know what you decide on Jackal as we are facing the same decision.
Adam Zacks-------------------------------------------Be Nice, Or Leave
October 16, 2008 at 9:51 am
No, dont say sorry.
The more discussion the better in my view. 🙂
October 16, 2008 at 9:58 am
🙂 Thanks 🙂
Adam Zacks-------------------------------------------Be Nice, Or Leave
Viewing 12 posts - 1 through 11 (of 11 total)
You must be logged in to reply to this topic. Login to reply
