| RE: [cosmos-dev] question on management domain and brokers |
|
Hi Hubert, Here is a sequence of events that will occur with a scenario of
2 Data Managers. We start with nothing running, and end up with a
consumer being able to query data from both of these Data managers. 1.
The Management Domain Starts (this is the entry point to the DC
world, lets just say we know where this is for now) 2.
The DataBroker Starts and informs the Management Domain where it
can be found (its EPR) 3.
DataManager_1 starts 4.
DataManager_1 asks the Management Domain “Please can you
tell me where the DataBroker is?” 5.
The Management Domain tells DataManager_1 where the DataBroker
is. 6.
DataManager_1 calls the DataBroker and says “Hello DataBroker,
My name is DataManager_1, and I can provide data on performance (for example),
and this is how anyone can call me... Voila! here is my EPR! 7.
DataManager_2 starts 8.
DataManager_2 asks the Management Domain “Please can you
tell me where the DataBroker is?” 9.
The Management Domain tells DataManager_2 where the DataBroker
is. 10.
DataManager_2 calls the DataBroker and says “Hello DataBroker,
My name is DataManager_2, and I can provide data on performance (for example),
and this is how anyone can call me... Voila! here is my EPR! 11.
The Consumer starts. 12.
Consumer asks the Management Domain “Please can you tell
me where the DataBroker is?” 13.
The Management Domain tells Consumer where the DataBroker is. 14.
The Consumer calls the DataBroker and asks “Hello
DataBroker, can you tell me if there are any Data Managers that can give me
performance data? 15.
The DataBroker looks up in its Yellow Pages all references to ‘performance’
and finds an EPR for DataManager_1 and DataManager_2 16.
The DataBroker informs the consumer of the EPRs 17.
The Consumer formulates (This formulation is a big subject that
needs to be refined and discussed at a later date) a query for both Data
Managers and sends the query to the appropriate Data Manager. 18.
The two Data Mangers take the query, and convert it into a
meaningful query to its API that does the retrieval of the data. 19.
The Data that comes back from each Data Managers’ API
calls is passed back to the Consumer 20.
The Consumer processes the data. 21.
If the consumer wants to get more refined data, it can go
directly back to the Data Manger, as, it already knows where it is. I hope that this clarifies things. Martin Simmonds CA martin.simmonds@xxxxxx From: cosmos-dev-bounces@xxxxxxxxxxx
[mailto:cosmos-dev-bounces@xxxxxxxxxxx] On Behalf Of Hawkins, Joel Hi Hubert, OK, here goes… >> I remember Martin said in the meeting that data managers
contact the management domain, not a broker. The management domain has a
UDDI registry, which is a logical place to start to discover location of
brokers. My recollection is that we decided NOT to make UDDI a requirement
for a DomainManager implementation. Regardless, the idea was that an external
client would go to the DomainManager to find out about all available
DataBrokers. If
data managers know how to contact brokers directly, the clients can do that as
well. Then what's the use of the management domain? From the point of view of a DataManager, I’m not sure there
is any need for a DomainManager. To me, it makes more sense to deploy
DataManagers with an affinity to one or more well-know brokers (well known from
the DataManager’s point of view) so that the DataManager can use WSDM
Advertisement to notify the DataManager of it’s availability, and let the
DataBroker roll up its knowledge of available DataManagers to the DomainManager
if one exists. From the client point of view, it seems easier to be able to
bootstrap use cases from a single point (the DomainManager), and possibly delegate
some of the client use case logic (for example, find me the DataManager for X)
to the DomainManager. What
I'm trying to clarify is not the "merged domain manager / broker"
simplified scenario, but the real multi-broker scenario. "Visit all
brokers until you find one that works" is not an elegant solution. Was that intended to be a smarmy comment? ;-) In any case, a client
will be “visiting brokers” one way or another. Taking your UDDI
case, each client would be forced to visit the known brokers in order to find
an appropriate match for a given request. If you interpose the DomainManager,
then it’s possible to offload some of that client implementation into the
DomainManager to simplify the marriage-making and allow for incremental optimization
without the need to update clients. Martin, Jimmy, Chris, please wade in! Cheers, Joel
Hi all, My understanding is that the
DataManager contacts its DataBroker, and the DomainManager only knows about
DataBrokers . The nice thing about this is we can easily support the ‘No
DomainManager’ simplified scenario. We can support routing either
naively (basically visiting each know DataBroker until we get one that works
for a particular request), or we can route based on prior knowledge (sticky
sessions and/or routing based on cached information about similar request).
This could be a place for ‘value-add’.
Good idea.
_______________________________________________ |
_______________________________________________ cosmos-dev mailing list cosmos-dev@xxxxxxxxxxx https://dev.eclipse.org/mailman/listinfo/cosmos-dev