I'm going to build a failover setup for my resque jobs system wich uses Redis as a queue. The idea is to put 3 redises in different availability zones on micro instances and replicate them. Also would be nice to use ELB/tcp for balancing and excluding failed instances(If its possible). Does any one has such experience? The main concern is that micro EC2 has inconsistent CPU, but it can be used for config servers on mongodb, so why not to put redis there? My queues aren't very big. Max amount of RAM consumed is around 40Mb and I have about 50 clients talk to the queue.
Amazon announced support for Multi-AZ support for Redis on 2014-10-24, so this you can achieve through the AWS console.
Regarding your other question about hosting your own EC2 instances with Redis, that is fully possible and will probably be cheaper. However, you will miss out on a lot of features (e.g. read replicas, automatic backups, detailed monitoring) and you will need to maintain it yourself.