RE: Active/Active Site A/Site B using SRDF

From: Matthew Zito <mzito_at_gridapp.com>
Date: Tue, 3 Apr 2007 17:41:49 -0400
Message-ID: <C0A5E31718FC064A91E9FD7BE2F081B1B97555@exchange.gridapp.com>

 

EMC SRDF does not support active/active with neither synchronous nor asynchronous configurations. When you have an SRDF environment your R1s (near side) are writeable while your R2s (the far side) are not.  

If you want to do active/active sites, you can do a RAC stretch cluster using either long distance Fibre channel (bleh) or iSCSI (much better), or you can do things at an application level.  

Matt

---

        From: oracle-l-bounce_at_freelists.org
[mailto:oracle-l-bounce_at_freelists.org] On Behalf Of ryan_gaffuri_at_comcast.net

	Sent: Tuesday, April 03, 2007 12:39 PM
	To: oracle-l_at_freelists.org
	Subject: Active/Active Site A/Site B using SRDF
	
	
	The SAN/Systems Administrators claim its possibly to have a Site

A/Site B setup(at two locations). Use SRDF asynchronously to populate and keep both sites active.          

1. I don't think its possible to actually write to sides(2 different databases) and still maintain transaction control? So at a minimum you write to just one site and then populate the secondary site.
2. If both sites are active and the population of the second site is asynchronous that implies that the second site will be slightly behind so if users query both sites then one user may get an inaccurate picture of the database.
3. Is it possible to have active/active with a synchronous SRDF? I would think that would affect performance. Since you can't end the transaction until both sides are applied.
4. I would think the better solution is to havea primary and failover with the load balancer having an exception handler so when Site A goes down, failover to site B.
5. If you want to use both sites to query, then you are better off identifying performance intensive queries such as reports and use the secondary site as a reporting database(unless Site A goes down, then site B handles everything)

        Even with all this you still have a single point of failure at your load balance since its the entry point or is there a way to multiplex this?          

        This is long... not sure how to summarize this.

--
http://www.freelists.org/webpage/oracle-l

Received on Tue Apr 03 2007 - 16:41:49 CDT