Understanding the Paypal use case architecture (no replies)

Hello,

I read the Paypal MySQL Cluster use case (http://www.itnews.com.au/news/twitter-paypal-reveal-database-performance-317811 and http://fr.slideshare.net/matkeep/paypal-mysql-cluster), but failed to grasp their architecture.

Paypal say they "scale by tiling", but does someone know how this is achieved? Is there some kind of load balancing between two "tiles"? In other terms, what is the nature of the link between two tiles? I understand they use geo-replication between datacenters, but within a datacenter, how do they achieve "tiling"?

Thanks in advance,
Best regards,
Pierre Sion