Understanding Flow of Globus Connect Server

[Joshua Brown]

Hi All, 

I am trying to understand the order in which certain processes are called in the Globus Connect Server version 5.4. The documentation provides this diagram.



What I don't understand:
1. How many different processors are running and what are their responsibilities, when I check the running processes on my system I see the globus connect server and globus-gridftp-server are there others?
2. Where does the storage connector/storage gateway fit into this picture, is this at step 7? and what process is responsible for using it?

3. What process is responsible for identity mapping? 

It would be useful to understand the architectural design of Globus Connect Server.

I apologize if there are better diagrams elsewhere in the documentation, I was unable to find them. 

[Jason Alt]

1. How many different processors are running and what are their responsibilities, when I check the running processes on my system I see the globus connect server and globus-gridftp-server are there others?

There are three logical processes that run on a GCS DTN:

 - GCS manager which services as the API endpoint and is the brains for GCS operations on the node

- The GridFTP server which is responsible for actual data transfer

- The GCS manager assistant which is responsible for keep the GCS node's configuration synchronized between other nodes in the endpoint and the encrypted configuration values stored in AWS

2. Where does the storage connector/storage gateway fit into this picture, is this at step 7? and what process is responsible for using it?

Connectors are loaded by the GridFTP server for directory listings and file transfers.You are right, that is step 7 in the diagram.

3. What process is responsible for identity mapping?

The GCS Manager process is responsible for identity mapping.

If you need any more details specific to your configuration, feel free to send mail to sup...@globus.org

Jason

[Joshua Brown]

Thanks for the response Jason, I still don't understand, according to what you said the connector i.e. POSIX, AWS... etc is handled by the GridFTP server which makes sense but to be clear that is separate from the storage gateway. When I create a storage gateway policy is that handled by the gcs manager as you allude to in your answer to question 3?

[Joshua Brown]

One more question. 

4. In the diagram posted above, the arrow associated with step 5 is confusing to me. When globus cloud application starts talking with the globus connect server it goes first through the GCS manager, not the GridFTP server? 

[Jason Alt]

Whenever you create or configure an endpoint/gateway/collection/role/domain, that is handled by the GCS Manager; it is responsible for managing the configuration for the endpoint. When you perform a transfer or directory listing (anything that uses GridFTP), the gridftp server receives the appropriate configuration needed to fulfill the request (which includes which connector to use and identity mapping) from the GCS Manager. GridFTP actually knows nothing (or very little) about gateways.

[Jason Alt]

The white 'globus connect server' box includes all processes we've talked about; gridftp is the only service listed explicitly. Transfer (globus cloud application) talks directly with the gridftp server; gridftp receives its configuration from the GCS Manager.