On Feb 7, 2017, at 11:09 AM, Amila Jayasekara 
<[email protected]<mailto:[email protected]>> wrote:

Sorry, mistakenly pressed "Send" button before I finish typing :-).

Oops I answered your first question in the other email :) Further replies below.

On Tue, Feb 7, 2017 at 11:01 AM, Amila Jayasekara 
<[email protected]<mailto:[email protected]>> wrote:
Hi Marcus,

Nice summary of requirements.

Some questions inline.


2. Have the ability to make calls to Identity Server (SOAP) and Airavata 
(Thrift).  Would be good if there is support for interacting with other remote 
computing protocols and HPC resources.

I was under the impression that IS has a REST interface also. Anyhow I am not 
sure about this.
Also, what you meant by "support for interacting with other remote computing 
protocols and HPC resources" ? Can you give few examples ?
IMO, if Portal need any information from the HPC resource it should go through 
Airavata for security reasons and also for communication reasons. If resource 
host a web server to extract information, it is find get information directly 
from resource (without going through Airavata)


I think you are right that in general those requests should go through 
Airavata.  Here are some examples I was thinking of:
* remote filesystem explorer over SFTP, GridFTP, Dropbox, etc. using the user’s 
own credentials
* pull in and display public catalogs of datasets


3. Be themable/skinnable and modular. Gateways should be able to customize the 
look and feel and make the portal their own. Gateways should be able to 
customize the functionality available in the portal, removing and adding just 
the functionality tailored for that specific gateway.

4. The backend framework should support different modes of web development: 
server-side rendered views as well as REST API for client side views.

5. The backend framework should be able to easily integrate with existing 
scientific codes for visualization and analysis. This would allow developing 
lightweight domain specific functionality within the portal itself.

6. The backend and frontend frameworks should have active communities and some 
longevity, helping to ensure continued support in the future.

7. The backend and frontend frameworks should have a gentle learning curve to 
promote onboarding new contributors. Setting up a local development environment 
should be relatively easy.

8. The frontend framework should allow the development of more sophisticated 
client side views:
        • Domain specific and interactive views for configuring applications 
and viewing results
        • Domain specific and interactive views for browsing, filtering and 
managing remote and personal data collections
        • More up front validation of application configurations



Proposed solution

With these requirements I feel like Django as a backend framework and React as 
a frontend framework are the best fit. Here are the benefits I see of adopting 
Django:
* Django has a builtin concept of modularity, the application [1].  This would 
allow the independent development of new views for new funtionality (req 1) as 
well as making it possible to tailor a portal to a gateway’s needs (req 3).
* Django has a very active community and lots of off the shelf applications [2] 
that can be used to either quickly develop common portal functionality or to 
add gateway specific features
* Being written in Python means that the backend can easily integrate with 
Python scientific codes for quick data analysis and visualization tasks
* A side benefit of a Python based portal is the opportunity to develop a first 
class Airavata client in Python. This could enable scripts to take advantage of 
the Airavata API or Juptyer like interactions with the API.
* Django is well-documented and relatively straightforward to work with, easing 
on-boarding of new developers

Regarding React, I see these benefits:
* React is a much more powerful and scalable approach to web UI development 
than just using jQuery, which is what the current PGA uses. With React we can 
build UIs with a higher degree of interactivity. And we can build richer, more 
domain-specific interfaces.
* Unlike other JS frameworks that have a very large API to learn, the concepts 
and API surface of React is fairly small. The main challenge is learning to 
“think in React” [3]

Again, feedback on this is much appreciated.

Its kind of hard to evaluate any product without actually prototyping. Django 
seems promising. However, I found this [4], on Django. Not sure whether those 
are really valid arguments.


Yeah, I think prototyping is important.  I’m currently working with Jun on just 
such a prototype for the GeoGateway.

I’ve seen similar complaints about Django in the past, but I’m not sure how 
much they still hold. That link you posted is almost 5 years old.  A lot of the 
complaints I’ve seen are about Django’s ORM, but the PGA would for the most 
part not be using that since data is persisted through the Airavata API.


[4] https://news.ycombinator.com/item?id=4076834



Thanks,

Marcus

Reply via email to