8-server advanced Domino clustering - can I do this?

I have a customer where we are considering going from 6 to 8 servers in the cluster. Either that or we will have two clusters consisting of 4 in one cluster and 4 in the other and then having scheduled replication of selected databases between the two.

We had some discussions today about what can be done with clusters. And I have a feeling that this scenario might work. I'll make a less complicated description of what we discussed below.

Servers:

A1, A2
B1, B2
C1, C2

B1 and B2 do not allow Notes RPC, only HTTP.

Let's say we have 100 databases in play across the 6 servers. 100 unique replica IDs.

On A1 and A2 we have 50 databases in replicas on both.
On C1 and C2 we also have 50 databases in replicas on the two.
On B1 and B2 we have all 100 databases in replicas on both servers.

A1 and A2 share 25 replicas with C1 and C2. In other words 25 replicas exist on all 6 servers.

The reason why there are some databases that we do not want available on all servers is that there is a huge amount of document creations and updates in these databases, leading to the Indexer task running all the time.
We will let the users from department A access databases on servers A1 and A2 and the useres from department B will access databases on servers B1 and B2.

We will be running this on Domino 8.5. The OS will be Windows 2003. It is most likely to be a 64 bit (both Domino and OS) setup.
The cluster will be running cluster replication on their own gigabit network.

Is this an effective and efficient setup?
Let's say that server A1 drops out. Will my Notes client try to access a database on a server that does not have it available?
When I create a document on server C1 in a database that only has replicas on C1, C2, B1 and B2 will C1 try to push the document to A1 and B2 too?

I have tried to google about these questions, but did not find any answers.


kommentarer

1 - -Let's say that server A1 drops out. Will my Notes client try to access a database on a server that does not have it available?

Its my understanding that notes will find the valid souces from cluster.ncf so it wont connect to a server that dont have a replica but my knowledge isnt up to date.
Skrevet af / Written by Flemming Riis fra 17:24:06 På 07-09-2009 | - Hjemmeside - |

Gravatar Image2 - Thanks a lot for the comments!

@2 Outside of support? Are we talking about the official IBM support here? Are you sure?

@3 B1 and B2 are serving up HTTP behind a Puakma ESSO Proxy { Link } .

Servers A1, B1 and C1 are in one server room, A2, B2 and C2 in a different one.
Both server rooms are in the same city.

No VmWare in play here.

Skrevet af / Written by Jens Bruntt fra 21:19:01 På 07-09-2009 | - Hjemmeside - |

Gravatar Image3 - @5 - I will check with IBM. Thanks!
Skrevet af / Written by Jens Bruntt fra 08:46:01 På 08-09-2009 | - Hjemmeside - |

Gravatar Image4 - Hi,

I guess it's difficult to describe such a complex set-up, because I quite don't get it:
50 unique db's on server A1+A2
50 unique db's on server C1+C2
and all 100 db's on B1+B2
But 25 of the 100 db's are on all servers !?!?
Hmmm..... Well, we leave it for now :)

Let me try to answer your questions.
Q: Is this an effective and efficient setup?
A: Hmmm.... my only concern is B1 & B2 ! These servers have all replica's but does not accept Notes connections only HTTP !
As you might know the LN client will try to find a database by searching on servers by name (first on servers starting with A then B and so on) - I sure cluster fail-over is working the same way.
I hope the real server name of B1 & B2 starts with a letter 'after' A1, A2, C1 and C2 :)

Q: Let's say that server A1 drops out. Will my Notes client try to access a database on a server that does not have it available?
A: Yes and no. The client will access the "first" available server in the cluster - in this case A2. If A2 doesn't have a replica of the requested database A2 will redirect the client to a server having the database. If the requested db is one of the 25 the client will try to open it on C1.
To redirect the client server A2 will use information in the 'Cluster Directory'.

Q:When I create a document on server C1 in a database that only has replicas on C1, C2, B1 and B2 will C1 try to push the document to A1 and B2 too?
A: I guess 'A1 and B2' is a typing error - should be A1 and A2 or ?? :)
No, server C1 will use 'Cluster Directory' information to decide where to the cluster-replication shall replicate the new document.

Splitting into 2 clusters:
If you have a performance problem this might help you. However you have to keep in mind you will only have fail-over in-side a cluster not across clusters. If you want fail-over between cluster you have to set-up: scheduled replication (with a very short interval) and some kind of IP-spray'er to 'switch' user access between the 2 clusters. I would stick to standard Domino set-up :)
Skrevet af / Written by Heinrich fra 07:58:51 På 09-09-2009 | - Hjemmeside - |

Gravatar Image5 - @7 Thanks. Yes, there are some inclarities in my description. But your answers make sense.
Skrevet af / Written by Jens Bruntt fra 08:02:24 På 09-09-2009 | - Hjemmeside - |

Gravatar Image6 - I have been seeing and hearing a lot of clients ask about this type of clustering.
Very MS in my mind.
If you NEED a specific db to always be available it really just needs to be on 2 servers, primary and secondary, if both are clustered.
Now due to geographical reasons you may need it on all 6-8, this also happens but you are balancing network performance/bandwidth with client side actions.
Like Paul said, alphabetical and added once found, but here is the issue.
If your local server is af01(africa) and bf01(backupafrica), if you have any other called am01(america), that will take the lead over bf01.
So planning is everything.
A bit over indulgent perhaps of a setup but if you really can not have a use b's server then you do what must be done.
Keep in mind, that for tertiary backup you may want the db's to reside on a 3rd server, "just in case" especially if your servers are all VM.
Another thing never put both your production and backup cluster servers on the same physical hardware box. Obvious, but it's been done, many times more than I want to recall.
What are you using for the B1/B2 http? an IP sprayer, if inotes? Or relying on Domino?
Mail redirection should also get reviewed to make sense.
Skrevet af / Written by Keith Brooks fra 18:47:39 På 07-09-2009 | - Hjemmeside - |

Gravatar Image7 - @2 - Im not actually 100% on it, but fairly certain that IBM supports a cluster only up to 6 servers, although you can have more. You may want to confirm that with IBM
Skrevet af / Written by Paul Mooney fra 21:35:08 På 07-09-2009 | - Hjemmeside - |

Gravatar Image8 - It will work - looks a bit over engineered to me, but it will work.

You do not need every database on every server in a cluster. The clustder directory will keep a record of all databases and where they exist.

Your notes client typically will look alphabetically at server names when looking for a replica after a server failure. When it finds one on a server, it will switch to it, write to the cluster.ncf file locally and then do it much quicker in future.

When you pass 6 servers in a cluster, you are outside of support though. I would go with 4 and 4, but as you know it really depends on your applications, geographical location and business
Skrevet af / Written by Paul Mooney fra 18:27:25 På 07-09-2009 | - Hjemmeside - |