I wonder how much of that he got almost-for-free just by using Terracotta?
Terracotta itself is in effect a transparent ACID object database where collections can be larger than RAM and objects can be accessed efficiently on one machine or shared and updated in implicit transactions.
Clustering and storage are completely based on Terracotta, and this is great: relying on a rock solid technology gives you the freedom to focus on more advanced features, such as communication, partitioning, data manipulation and querying.
Which is the same reason Voldemort and Riak rely on third-party storage, or HBase relies on Hadoop.
In fact Terastore is based on Teracotta is one of the most important aspects. This makes us comfortable to see similar performance in numbers related to clustering, storage. Same is true for MemchacheDB, Voldemort and Riak as Sergio mentioned.
Exactly: clustering and storage are more and more "commodity" features, as I see it.
The most important thing is going to be what you build upon them, the real value you provide to the user: is data partitioned or replicated? What data manipulation features are provided? And so on ...
So like, all of the hard stuff. Implementing a protocol wrapper around Terracotta should be cake. Of course, it does force the end user to learn how to administer Terracotta, which non-trivial in it's own right.
Comments
I wonder how much of that he got almost-for-free just by using Terracotta?
Terracotta itself is in effect a transparent ACID object database where collections can be larger than RAM and objects can be accessed efficiently on one machine or shared and updated in implicit transactions.
Clustering and storage are completely based on Terracotta, and this is great: relying on a rock solid technology gives you the freedom to focus on more advanced features, such as communication, partitioning, data manipulation and querying.
Which is the same reason Voldemort and Riak rely on third-party storage, or HBase relies on Hadoop.
In fact Terastore is based on Teracotta is one of the most important aspects. This makes us comfortable to see similar performance in numbers related to clustering, storage. Same is true for MemchacheDB, Voldemort and Riak as Sergio mentioned.
Exactly: clustering and storage are more and more "commodity" features, as I see it. The most important thing is going to be what you build upon them, the real value you provide to the user: is data partitioned or replicated? What data manipulation features are provided? And so on ...
So like, all of the hard stuff. Implementing a protocol wrapper around Terracotta should be cake. Of course, it does force the end user to learn how to administer Terracotta, which non-trivial in it's own right.