[Seaside] Enterprise Software V Ecommerce [in
Smalltalk at JGFoster.net
Thu Feb 23 00:07:07 UTC 2012
It is certainly true that having a relatively clean separation between the layers (infrastructure, domain, application, and UI) is a good design principle. And Steve is right that building a complex network of objects that cannot fit in memory is not very scalable.
On the other hand, if you have a tool that allows you to act like you have a huge transactional object space that is shared by hundreds of concurrent sessions, where your "stored procedures" are written in Smalltalk, and Seaside support is built-in, then your prototype might not be so unrealistic. Have you heard of GemStone/S? Check out http://seaside.gemstone.com/, and the various links from there, including my blog at http://programminggems.wordpress.com/ and Dale's blog at http://gemstonesoup.wordpress.com/.
Feel free to ask more detailed questions...
On Feb 22, 2012, at 3:33 PM, dirk newbold wrote:
> Hi all,
> I’m a relative newbie (4 years) to programming and seaside, who
> partnered up with some “old hands” (smalltalk/seaside is my first and
> only language).
> Initially we developed an Ecommerce (Istazaar) fashion platform for
> designers to maintain their collections. We stagnated last year due
> to the time restraints on the “old hands” and I then began developing
> a web based Enterprise Software (Vivport) platform for
> We now have two ventures at relatively equal stages of development.
> I believe the Enterprise Software (Vivport) platform is the more
> innovative however, now that Steve (one of the “old hands”) has some
> time again he has been able to review its development (which
> unfortunately I built unsupervised) and has major concerns in relation
> to deploying this venture.
> Instead of building the database; then building middleware, then UI;
> I’ve basically just built a UI, Steve’s concerns with the outcomes of
> this development process are outlined in his email to me below.
> We were wondering if the smalltalk/seaside community has any views on
> Steve’s concerns below that may influence our decision on which
> venture to proceed with.
> The platform has been developed to freely change between the user’s
> personal account and their company(ies) account. The personal
> accounts are free so we would be attempting to server many users.
> Any advice in relation to users/image, persistence, database, general
> deployment etc for this type of venture (essentially a beefed up
> gmail/cloud environment) would be greatly appreciated.
> Thanks for your time.
> ---------- Forwarded message ----------
> From: Steven Berryman <Steven at berrymanelectrical.co.uk>
> Date: Sat, Feb 18, 2012 at 6:07 AM
> Subject: RE: Vivport V Istazaar
> To: dirk newbold <dirkdirk at gmail.com>
> As mentioned I’m concerned with the approach you have taken with
> Vivport and in one way this technology can lead you into a false sense
> of how easy things are (trust me you couldn’t have done it in .NET or
> java without infrastructure). The difference with the ecommerce site
> is that there is little shared data and I knew that the only real
> issue was storing users but that would be a simple mapping issue to a
> table. Even the registered objects were designed to make storing in
> the database easier so objects didn’t directly hold onto a registered
> object. What you have built is a very complicated network of objects
> for the benefit of the UI which for a prototype is great but would
> never work in the real world. The type of project you are building
> with Vivport is infrastructure and in one way the last thing you
> should have done is the UI. One system I worked on had 3-4 months of
> designing the infrastructure before a line of code was written.
> The problem you have is that you have assumed that all objects would
> be in memory and are accessible via threads and this just isn’t
> possible. If I look at my local Microsoft exchange server it’s
> currently 50GB for 5 users and the machine only has 4GB of ram so not
> only do you need to deal with large amounts of data but you have not
> addressed any multi-threading issues. For example Smalltalk provides
> structures such as SharedQueue to deal with any shared data in the
> memory and I’m guessing you didn’t use these? When everything is in
> memory the world is very easy but for a system like yours you would
> need to start with the database and design the schema. Most searches
> would be done directly in sql (assuming you are using a relational
> database) so that way you reduce the amount of data going to the
> client. So when you open up Microsoft Outlook it will run a query on
> the database to get your inbox and the email headers so you don’t pull
> large amounts of data across the network. As you start to click on
> emails it will pull the data over as required.
> So here is an example to show multi-threading issues (I have not used
> your code but I’m sure you get the point)
> Lets’ assume you have a folder with 10 items
> Dirk - actions Danny - actions
> folder items do: [ :item
> “goes through this loop twice”
> delete the last
> item in the folder
> “code falls over because the last element
> in the collection does not exist.” ]
> I’m sure you could try something like the above but you would need to
> put in a break point in the Dirk loop so you could then go in and to
> the removal from a different session. Now imagine the same thing with
> 5,000 users logged into your server all running threads across your
> code. Your code would have to be thread safe and this is not a simple
> task especially trying to do it on code not designed that way from the
> beginning. You could end up locking up all the other sessions if you
> don’t carefully design and keep your critical sections to a minimum.
> If you are using collections that are hashed they will fail in very
> strange ways if you have other threads removing or adding items as you
> are iterating through the collection. Basically in the ecommerce site
> we could sort of ignore the shared data issue as we shared very little
> data between sessions.
> Also you have added features that would be very expensive when using
> databases. For example the running man relies on the status of some
> piece of data and of course when everything is in the same memory
> space this is easy. In a database you would have to cross a
> transaction boundary and even the fastest database servers we have
> (costing 200k per server) you would be lucky for a single transaction
> to take between 1-10 milliseconds (costly when you scale up users and
> make too many database calls). This is why Microsoft exchange works
> the way it does as you can imagine how slow it would be if 5000 people
> logged into the server at the same time. So for most document
> management systems they would have a large database at the back end
> and all the searches would be done as stored procedures that could be
> based on some proprietary algorithm and none of this would be done in
> the language itself such as Smalltalk. Until you have a database
> schema designed you can’t even think about the UI.
> If you want to prove to yourself then try and create 100,000 users all
> with about 10MB of data and test to see if it works. See if you can
> log in a few sessions and send emails around.
> With the ecommerce site I could imagine how to add a database at the
> backend because it would be very simple. I would have no idea how you
> get this into a database even if you used an object database you would
> still need a defined object model. It’s also the reason why I liked
> the ecommerce site as it would be simple from the database end as it
> would only contain users, basic static data and designer’s data. Also
> I can see how we could have multiple servers to handler more traffic
> with the ecommerce site but would have no idea how we could do the
> same with your design.
> Hope this makes sense. Let’s catch up next week.
> seaside mailing list
> seaside at lists.squeakfoundation.org
More information about the seaside