Tuesday, March 31, 2009

Production e-infrastructure

There was a recent (heated) discussion within the Production Grid Infrastructure Working Group within the Open Grid Forum with a thread started by Laurence Field.

In reality, what resource providers would class as a production grid, most users are beginning to treat as e-infrastructure. So, what are the characteristics define e-infrastructure?

1. They can span multiple organisations
Physical resources (e.g. computers, storage, instruments, data, ...) that are owned by organisations are made available to collaborators within the same organisation (e.g. within a single legal entity such as campus), between different organisations (but within the same legal jurisdiction such as a country) or internationally across legal boundaries. The potential span of the collaborators within the e-infrastructure has a direct impact on the policies governing the e-infrastructure.

2. They allow teams to collaborate
Any e-infrastructure must be able to support individuals from different organisations to work together in teams (or virtual organisations). These teams should be easy to form and be able to support roles/groups within the team. Access to the resources within the e-infrastructure should be governed by the identity or role/group and individual has within a particular team.

3. They have a scalable decentralised authentication scheme
For the uptake of e-infrastructure to scale an individual must be able to obtain locally an identification token that can be used to identify themselves to remote resources. Currently, many deployments use X509 certificates rooted in Certificate Authorities issued from the International Grid Trust Federation (IGTF) to identify an individual. These are then frequently used to obtain an X509 Proxy Certificate (as a means of encapsulating an identity with additional information that can be used for authentication) from VOMS. Alternatively, institutional based authentication schemes such as Shibboleth are beginning to see deployments.

4. They have a reliable scalable service based infrastructure
Service oriented architectures (SOA) have become the defacto standard model for building distributed computing infrastructures, regardless as to if they are implemented using web service specifications or some other technology. By definition, a SOA needs a scalable information system that allows services to publish information relating to their capability and interface, and mechanisms by which clients can discover such information and select the desired service(s). The underlying infrastructure needs to be reliable and able to scale to meet the needs of its service providers, the number of services within the system, its user communities, and its workload.

Perhaps the biggest difference here and earlier 'grid' definitions is the focus in e-infrastructure on reliability and scalability.

No comments: