Some of the nodes on the us-east-1 load balancing cluster suffered a severe network partition causing some requests to fail. The faulty nodes have been replaced in other working AZ's and we're seeing the number of errors decreasing.