In preparing for my Web 2.0 Expo presentation “Forecast: Partly Cloudy,” I have been looking at how existing code can be reused in the cloud.  There are several big challenges to reusing existing code in the cloud:

1. Many cloud platforms support only one specific language (e.g. Python for Google App Engine, Ruby for EngineYard).  That excludes a lot of code written in PHP, Perl, Java and other languages.
   

2. Almost all cloud platforms have a non-SQL data access model. This matters for code that has explicit SQL statements in it, which is unlikely to ever be meaningfully available anywhere (SQL after all is one of the issues with the existing stack).
   

3. A lot of code uses libraries written in C or C++.  Support for these libraries will be take time to implement on many cloud platforms as they cannot sandbox the existing code.  In fact, some of the libraries will never be supportable because they permit low level access that cannot / should not be reproduced in a sandbox.
   

4. Letting code in different languages talk to each other is a non-trivial problem.  Google internally uses something call Google Protocol Buffers which they have opensourced.  Facebook has a similar solution called Thrift (also open sourced).  Both of these are IDL-based code generators, which require effort in defining interfaces, generating stubs and connecting them up to existing code.

Our portfolio company 10gen is addressing all of these issues.  First, they are intentionally designing the 10gen application server to support multiple languages.  They have already announced Python support (in addition to Javascript) and are working on several other languages.  Second, while 10gen’s Mongo DB does not support SQL access, it has a rich query language that makes it easy for 10gen to support frameworks such as Ruby on Rails where data access is model-based.  Third, 10gen is open source and is leveraging existing open source language projects that target the JVM, such as Jython.  This means that as those projects add Java-based support for a specific library that library will also be available on the 10gen platform.  Fourth, 10gen is building a bridge layer into the appserver that lets existing code be invoked from other languages without the need for an IDL and code generation.  On 10gen it will eventually be possible for Python code to grab a PHP object and invoke a method on it.  All of this is exciting because it greatly facilitates the ability to reuse existing code in the cloud and as such the adoption of cloud computing.

In preparing for my Web 2.0 Expo presentation “Forecast: Partly Cloudy,” I have been looking at how existing code can be reused in the cloud.  There are several big challenges to reusing existing code in the cloud:

1. Many cloud platforms support only one specific language (e.g. Python for Google App Engine, Ruby for EngineYard).  That excludes a lot of code written in PHP, Perl, Java and other languages.
   

2. Almost all cloud platforms have a non-SQL data access model. This matters for code that has explicit SQL statements in it, which is unlikely to ever be meaningfully available anywhere (SQL after all is one of the issues with the existing stack).
   

3. A lot of code uses libraries written in C or C++.  Support for these libraries will be take time to implement on many cloud platforms as they cannot sandbox the existing code.  In fact, some of the libraries will never be supportable because they permit low level access that cannot / should not be reproduced in a sandbox.
   

4. Letting code in different languages talk to each other is a non-trivial problem.  Google internally uses something call Google Protocol Buffers which they have opensourced.  Facebook has a similar solution called Thrift (also open sourced).  Both of these are IDL-based code generators, which require effort in defining interfaces, generating stubs and connecting them up to existing code.

Our portfolio company 10gen is addressing all of these issues.  First, they are intentionally designing the 10gen application server to support multiple languages.  They have already announced Python support (in addition to Javascript) and are working on several other languages.  Second, while 10gen’s Mongo DB does not support SQL access, it has a rich query language that makes it easy for 10gen to support frameworks such as Ruby on Rails where data access is model-based.  Third, 10gen is open source and is leveraging existing open source language projects that target the JVM, such as Jython.  This means that as those projects add Java-based support for a specific library that library will also be available on the 10gen platform.  Fourth, 10gen is building a bridge layer into the appserver that lets existing code be invoked from other languages without the need for an IDL and code generation.  On 10gen it will eventually be possible for Python code to grab a PHP object and invoke a method on it.  All of this is exciting because it greatly facilitates the ability to reuse existing code in the cloud and as such the adoption of cloud computing.