Does anyone have experience with Scalearc sharding and it's ability to scale ? Does Scalearc needs to have the whole data in memory or does it behave well with disk storage ? I'm thinking to big database application like 10TB and above.
ScaleArc does not do Data Sharding
I actually wrote an earlier post where I evaluated it for my former employer : Experience using ScaleArc in test or production?
I doubt that you have the budget for multiple servers using 10TB of disk each.
ScaleArc is primarily a read/write splitting solution, sometimes called clustering. To the best of my knowledge they don't do sharding.
Each 'node' has a complete copy of the data and queries are redirected to one or the other 'node'. The challenge with this primarily that you have many nodes, each with a whole data set, and you only get to accelerate 'read' style queries, not 'writes'.
For the kind of volumes you are looking at, a parallel database architecture seems more appropriate. They would scale considerably better with that kind of data volume and give you very high performance access to all of the data irrespective of the location of a given row of data (i.e. which 'shard' it resides on). Parallel database architectures are considerably superior to 'sharding' and you may want to review some literature such as this paper my Dewitt and Gray.
I notice in your comment to another question that you mention that you are evaluating other solutions. Of the solutions you list, ParElastic is the only one that is truly a 'parallel database' architecture. Full disclosure, I work for ParElastic, and I only mention this distinction because you indicated that you were evaluating these products in your earlier comment.