perfsonar development work

["Jeff W. Boote" <>]

Vedrin Jeliazkov wrote:
> Hi Jeff, Joe, Maciej, All,
>
> "Jeff W. Boote" 
> <>
>  wrote:
>
> <snip>
>
> > Suggestions from GUI developers about what types of queries they intend to
>
> make
>
> > (what information they need) would be very helpful. I just have not been 
> > thinking along these lines lately.
>
>
> There might be several different queries, as described below (in increasing
> order of complexity and expected performance impact);
>
> 1) get the addresses of services, registered in the LS;
> 2) get the addresses of services, registered in the LS and having a given type
> (e.g. utilization MA, Hades MA, etc.);
> 3) get the addresses of services, registered in the LS, having a given type
> and providing info for a given data point (e.g. IP interface address);
> 4) get the full list of data points for a given service;
> 5) get the full list of data points for all registered services;
>
> Query (1) is mandatory. Query (2) would be perhaps sufficient for making the
> current MA.conf in perfsonarUI obsolete. Query (3) would greatly simplify (and
> probably make faster) path visualization, because the clients would not have
> to query a range of other services for each successive hop. Queries (4) and
> (5) might be desirable in some cases (e.g. when drawing maps), but perhaps
> they could be run much less often, because of the expected performance impact
> on the LS itself.

I realize this is not reasonable for an immediate fix, but I did want to bring 
up something else we have talked about in the past.

The MA should have the functionality to respond to LS message queries that are 
specific to it. For example, the MA itself should be able to answer the question 
about what IP interface addresses it is holding data for. I'm not convinced it 
is a good idea for each and every interface address to be loaded to the LS. 
There should be aggregation involved here.

If this were available, it would be fairly easy to do the first 3 of Vedrin's 
queries. 3 would need to be somewhat fuzzy - you ask for data for IP address X 
and the LS needs to match X against the aggregation function it is using. Most 
likely a netmask for IP addresses. The client could then interrogate the MA 
directly to see if it has data for the specific IP address.

4 should be directed to the individual service.
5 should be the client accessing each individual service to make the query.

jeff