Need more details :-). What are you trying to do?
Right now PI Notifications failover with an AF Collective does not work when the Notification schedulers run on the AF Servers. You know because Notifications decides to write to the configuration AF Database to say which Notification scheduler is the primary during Notifications failover, but you can't if your AF Collective has failed over to the secondary.
Assuming abacus has a similar distributed scheduler based on the Notifications architecture then we wouldn't want to hit this same problem with an AF Collective.
PI AF Server will need a clustered or mirrored SQL Server even with an AFCollective. An AFCollective, is intended for scaling out of read-only access to AF metadata. If you require an AF deployment that cannot tolerate the loss of the primary SQL Server, it is necessary to deploy a clustered or mirrored SQL Server and deploy the AF Server service on a load balanced web farm or on a cluster. In the future, SQL Server 2012 Always-On will likely be supported by PI AF Server.