I have a few questions about IssueVision (from WindowsForms) concerning its
scalability and performance. Rather, if I were to implement techniques
described here into another application, how would it perform, how well will
it scale, and what considerations should I be take into account?
First, let me say that I appreciate the work done to get this application
out there for me to look over. I continually refer to it and I think I am
ready to design an application using this model. Without samples like this,
I would not be enlightened enough to ask more question, so, thank you.
Let’s say we have the proverbial order entry application, in this case we
sell products and services. Each client application would need the ability
to work in a disconnected mode from time to time.
The application flow would resemble this:
Sales Rep Prepares a Quote
- Quote Header
- Date/Time
- Sales Rep ID
- Customer Information
- Billing Address
- Mailing Address
- Quote Details
- Product Kits (Grouping of Parts)
- Parts
- Services (Installation)
Sales Rep Converts a Quote to Sale (Workorder and Invoice)
- Order Header
- Date/Time
- Sales Rep ID
- Customer Information
- Billing Address
- Mailing Address
- Payment Terms
- Order Details
- Product Kits (Grouping of Parts)
- Parts
- Services (Installation)
Data in Synchronized with Central Data Store
Fulfillment Rep processes the order
- Opens Workorder Ticket
- Locates and freezes parts
- Backorder status
- Warehouse transfer request
- Check Service Rep Availability
- Assign Technician
- Monitor Workorder status
Billing Processes Invoice / Payments
Sales Rep Monitors Order Activity for this Sale
While Finance and Fulfillment would most likely be connected to the server
at all times, the sales force will be required to host a lot of data locally.
What’s more, it would seem that the sales rep would have to load a lot of
data into memory at one time.
1. Is there a way to minimize how much data is loaded into memory at a time
and still use Subject/Observer and Disconnected methods?
2. According to the documentation on Microsoft’s site about the
Model-Controller-View model, there is a propensity to code memory leaks into
the system, how do I avoid them? How do I test for them?
3. It would seem that to convert the GUI to a web application, it would
require some major code changes to make it work. For example:
a. Helper functions (While not required for a web client) use the
Application namespace, and therefore could not be included into a generic
assembly)
b. The only piece that you could isolate into a common assembly would be the
Controller (IssueSubject) and the patterns (IObserver, ISubject).
It appears as though the goal in IssueVision was to make a narrow data
model. I think I understand why it was designed that way, first of because
it is efficient, and second this was to demonstrate technologies in use.
However, in the case of our ordering system, our Orders table would have two
or three foreign keys (depending on how you model your data) Shipping
Address, Billing Address, and Customer. I have a few questions concerning
this:
1. Using the IssueVision model, how would one implement Transactions? I
would create a transaction object in my Version of the IVData object and
create a transaction as I needed. Would I still be informed of conflicts as
they occur and would they be only for when the client/server copies differ?
2. Using the dataadapter’s Update() method and handling the row updated
event for addresses, how would I know which address row represents the
mailing address and which represents the billing address (Assuming the
addressed is an autonumber int)
3. Using the IVData component implicitly instantiates all the data adapters
and command objects at once. Is there an overhead associated with that, is
there a better way? If I were to instantiate the objects that I need as I
need them, do I introduce other issues or affect performance?
4. Would it be more efficient to send/receive the data serialized than to
pass a dataset?
When loading the data, I can easily see that the parts catalog would be
lookup reference data. I can further see that this data only needs to be
updated on demand or periodically and not every time the application is
launched. I think I get that well enough.
When loading the open orders though, I have a question about the Orders
foreign key partner tables. In the case of addresses I would not need to
load that data while displaying the information in the data list. I would
need to load that data while working with the individual item though. Enter
the dataview and my next set of questions:
1. First, is there a better way then dataviews?
2. How do I cast a datarowview back to table.rows()?
3. When looking up the shipping address, I assume I would need to extend the
subject object to include a method for returning datarowview (dataview seems
to be the only object to allow filtering of loaded data), when updating this
information is the data still a member of the dataset?
When creating an application a bit more complex than IssueVision, what are
some considerations to take into account, pitfalls of development (besides
debugging a webservice), and finally is this the right model to implement for
an application larger than IssueVision?
Thank you for your time.