Exchange 2010 DAG - Failover selection between sites

  • Thread starter Gunja69
  • Start date Views 930
Status
Not open for further replies.
G

Gunja69



Hi

I would like to get some clarification on the following scenario.

3 sites with 2 mailbox servers at each site. 6 servers in total
All 6 servers are members of 1 DAG.

I want high availability and site resilience. The question is this.

Can I have these servers in one DAG but only have automatic failovers between servers locally in the same site and only do manual switchovers between sites?
 
N

Neil Hobson [MVP]

I don't think so. You'd probably need 3 DAGs (one for each site) and for each DAG have 2 members in one site and a third member in one of the other two sites. The third member could be a lagged database copy that is also marked as 'activation blocked'. In other words, each site has two servers operating normally and therefore giving high availabilty, but there is another copy of the data at another site that will not be chosen by active manager and used unless the administrator intervenes.Neil Hobson, Exchange MVP
 
B

Brian Day MCITP [MVP]

Neil, if he uses activation blocking I think he can get what he wants with 1 DAG. I'm assuming the setup would be something similar like this...

Site-1
Server1: DB1, DB2, DB5
Server2: DB1, DB2, DB6

Site-2
Server3: DB3, DB4, DB1
Server4: DB3, DB4, DB2

Site-3
Server5: DB5, DB6, DB3
Server6: DB5, DB6, DB4

And the file share witness is somewhere, but not shown above.

If the last DB copy up above for each server is activation blocked then intra-site failover should still work and inter-site failovr would require manual intervention. Also I believe (will have to double check) that a DAG will always prefer the lowest DB preferred activation # in-site before crossing site boundaries.Brian Day, Overall Exchange & AD Geek
MCSA 2000/2003, CCNA
MCTS: Microsoft Exchange Server 2010 Configuration
LMNOP
 
N

Neil Hobson [MVP]

Brian - of course - you're right. I didn't think it through fully before replying. That'll teach me. :)

I think the subject of lagged copies might be worth looking at here for the inter-site copies. Does that have a bearing on activation outside of blocking? I can't remember without checking.Neil Hobson, Exchange MVP
 
B

Brian Day MCITP [MVP]

I can't remember either without going to look. I think lagged copies are not activation blocked by default (I could easily be wrong here), but it is suggested that they should be blocked since you'll have to wait for log playback to complete.Brian Day, Overall Exchange & AD Geek
MCSA 2000/2003, CCNA
MCTS: Microsoft Exchange Server 2010 Configuration
LMNOP
 
N

Neil Hobson [MVP]

Good point about the log replay - could take quite some time depending on the length of the lag. In relation to the original question, you'll be pleased to hear that I've made a strong mental note to remember that individual databases can be activation blocked as well as via Set-MailboxServer. To think someone here in the office asked me some Exchange 5.5 questions recently.....I can't keep up with all the questions on the various versions. :)Neil Hobson, Exchange MVP
 
M

Martin Chisholm -

Scott wrote up a good description of the current implementation of Best Copy Selection (BCS).
http://technet.microsoft.com/en-us/library/dd776123.aspx

By setting a replay lag of >50 logs, you can artificially induce a lower priority for BCS.

And as always, we reserve the right to change the implementation details in future versions (including service packs).

Neil: I hear you. It's much easier if you can forget those older versions, but I realize not everyone has that luxury. :)

-martinLegal Stuff: This posting is provided "AS IS" with no warranties, and confers no rights.
 
B

Brian Day MCITP [MVP]

Scott wrote up a good description of the current implementation of Best Copy Selection (BCS).
http://technet.microsoft.com/en-us/library/dd776123.aspx

Martin, I have a couple questions (I know... shocking!)

In that article in the Database Failover section I don't quite understand what the ACLL is doing the way it is currently worded. When a DB failure happens is the ACLL looking to see what the highest log *any* DAG member with a copy of that DB had and inspected before the active copy failed and if so copy all missing logs to every other node if they didn't yet get them? Will it then choose what other DB copy to mount through the selection logic? If one server is unavailable to query, is that considered a failed ACLL and it goes straight to AutoDatabaseMountDial value, then selection logic? Sorry, I just couldn't quite follow that paragraph well. Maybe I need sleep.

Also, are in-site copies (the same site as the copy that just failed) always preferred even if the lowest ActivationPreference is for some odd reason in a remote site at that time, or will it absolutely always assuming all other copies meet the same selection criteria go with the lowest value even if in a remote site?

Thanks!

-brianBrian Day, Overall Exchange & AD Geek
MCSA 2000/2003, CCNA
MCTS: Microsoft Exchange Server 2010 Configuration
LMNOP
 
Status
Not open for further replies.
Top