The Grand Cloud

We have posted a new article about the protection of Intellectual Property in Romania.  We believe the protection of Intellectual Property is a critical part of off-shore development.

Joa Ebert posted a very interesting blog post about Compiling ActionScript in The Enterprise and I thought it would be interesting to explain what I did a while ago as part of the Anvil Flex Open Source Project to speed up compiling Flex 3 applications.   I should say we, because I had a lot of help from James Ward in building this.  Anvil Flex was desinged somewhat like a portal server and the application was then broken down into smaller sub-applications or "modules" (in the orginizational context and not the flex context).   Each module was designed so the it could be compiled and tested in a stand alone context.  There was also several common libraries that where shared with the entire application

About 6 months into this project compile times for the entire application where exceding 20 minutes and it was nearly impossible to do continuous integration.  One of the main problems with the flex compilers is each part of the compilation process (mxmlc, compc, opimizer, etc.) requires you to start a separate JVM, compile that small part of code and then shutdown the JVM.  Not only does starting and stoping the JVM take time, you also lose all the compilation information when you shut down. 

To solve this problem we build a custom Java compilation program called Taz that called directly into the Flex compiler classes, since the Flex compiler is written in Java.  It consited of the following steps:

1 - We designed a heirarchial build system in Ant that would allow you to build any part of the tree and it's descendets automatically.

2 - The common libraries and modules each had there own Ant build files that described their dependecies.

3 - On building the application we would first walk the tree and build a dependcy heirarchy in memory of what modules and libraries need to be built.

4 - Then as we compiled each library we would cache any pre-processing and loaded libraries that could be reused throughout the compilation process.

The result was compile times dropped from 20 plus minutes to around 4 minutes!    Then for continuous integration builds we set-up multiple builds that would first compile the individual modules and common libraries and then build the entire application.  This would prevent a single module from breaking the build of all the other modules.

A very interesting and logic next step would be to take this process and allow the compilation process on individual developers machines to use the latest libraries built remotely.  For example if a developer was working on module A that depended on module B they could specify to grab latest pre-built module B and any common libraries off a remote machine, like the last known good build from the continuous integration server.  You could also distribute the build across machines this way.

A final note,  a lot of people don't like Ant and say Maven is the build system of the future.  Acutally part of the reason I stop working on Anvil is the entire build process was in Ant and I also thought the trend was to using Maven for builds.  Then on my last project we used Maven extensively and I know believe that Ant actually might be better solution for compiling large and complicated Flex Applications.  One of the reasons I didn't like the Maven build system was a large part of our build process actually used Ant behind the scenes for everything that could not be easily done in Maven.  Yes, Flex Mojos and such can help compile a fairly standard Flex project, but when you project reaches a certain point of complexity that starts to fall apart.  Things such as bundling custom resource files, including special assets in the final build, linking common library files, etc.  it gets complicated really fast! 

I had previously said that the Anvil Flex project was dead, but the code lives on in sourceforge. If there is any interested this build process could be extracted out into a separate, stand alone project.

 

The Flex SDK team finally decided to fix the problem with Adobe Air Auto-Update in Flex SDK version 3.5a.  Here are the notes from QA on the bug:

The Flex team has updated the 3.5 SDK with a small update that addresses this issue. The refreshed build, SDK 3.5a, has only a few files modified in order to fix this issue and this change does not affect the signing and caching of the SDK 3.5 RSLs originally released in December.

We encourage all developers using SDK 3.5 to upgrade their build to SDK 3.5a to continue their development. The SDK 3.5a build can be found in the “Latest Milestone Release Build” table here: http://opensource.adobe.com/wiki/display/flexsdk/Download+Flex+3

Thank you for your patience as we worked to address this issue.

After spending a long time yesterday trying to get the Adobe Air Auto-Update to work I realized it was a bug with Flex SDK 3.5.  What happens under Flex SDK 3.5 when you run the auto-update is you only see a blank / gray screen pop-up with no wording.   At first I thought it was my code so I tried it using Greg Wilson's sample.  Compiled under Flex SDK 3.5 it doesn't work at all, but then I tried it under 3.4 it works great. I have filed a bug with Adobe, so please vote for it so we can get a quick fix -http://bugs.adobe.com/jira/browse/SDK-24919

I decided to to file a number of bugs for issues we have been seeing with flex builder.  I have posted them here in hopes you will go vote for them.  This is only some of the many issues we are seeing using Flash Builder.

Allow multiple instances of an AIR app to run at the same time. - There was a previous issue for this which was closed, so I created a new issue with more details.

Need to be able to navigate to classes and methods that implement an interface

Flex Server ate my bin! - This bug was already in the system, but looks like it is stuck on waiting on Info -  Hopefully voting on it will get it moving.

After changing a project type from Flex Server I still can not set my output folder in Flex Build Path

Multiple runnable Air applications in the same project using locale causes tabs display name to be null

Should be able to navigate into script source with Go to Definition Looking through the bug database it looks like several of the renaming bugs have recently been fixed, so maybe renaming will work in the latest release.

I had a terrible bike accident that tuned into a major ordeal. I was coming down the hill way to fast on my bike when I fell off the trail in a patch of soft dirt and went flying over the handlebars. I was clipped into my pedals and did not come out in time to catch my self. I ended up taking the entire crash in the face (I am not sure what I did with my hands). What happened after was worst then the accident itself.  I was treated in two regional hospitals that almost killed me.  The my life was saved at a third hospital.   Sadly I found you can not trust our modern medical establishment.  What I learned after is every nurse and doctor should know the ABC's of keeping someone alive - airway, breathing and circulation.  What is scary is two of the hospitals I visited almost killed me by ignore this basic principal.  The third hospital realize I had a severe infection from the fall and my airways were caving in.  I had to be put on a respirator tube and kept in the ICU for a while.

I put together a cloud hosting comparison chart on Zoho DB.   You can also view the full DB and some reports on Zoho DB.  I need to figure out better ways to visualize the data, so let me know if you have any ideas.  I had to makesome assumptions to try and make the comparisons somewhat equal. Let me know what you think.

Mosso just announced they are going to be offering cloud servers at $10.95 / Month.  That is about half of the cost of of the competition.  This cloud service is based off their acquisition last year of Slicehost.   It will be interesting to see how similar this offering is to Slicehost.  They have a good reputation for excellent customer service, so hopefully they will keep this up. What else was interesting was they have updated Jungle Disk to support Cloud Files.  The cost currently only includes the monthly storage cost and there is no cost for data transfer.  That would make the price much lower than storing files in Amazon S3 through Jungle Disk.  The cost for direct use of Cloud Files though does include fees for data transfer. Update: I just realized that the cost of Cloud Servers does not include data transfer.  They charge $0.08 / GB for bandwidth in and $0.22 / GB transfer out.  If you assumed 50 GB in and 100 GB out that would add $26 to the total cost.  So it is not as competitive as I originally thought.

Amazon today announced  EC2 Reserved Instances.  This allows companies to pay a one-time fee upfront for a guaranteed reserved instance: Amazon Web Services Blog: Announcing Amazon EC2 Reserved Instances This brings the price down of using EC2 for hosting a web site, similar to using a VPS.  For comparison we put together a chart showing how the price compares to other major VPS providers.  We made a number of assumptions to calculate the monthly price for Amazon EC2.  I have also published  more comparisons on this Google Doc.

Hosting Provider	Monthly Cost	Ram	Storage	Transfer
Amazon Reserved Instance *	$70.68	1741MB	160GB	50 GB Data In / 100 GB Data Out
Slicehost	$70	1024MB	40GB	400GB
Linode	$59.95	1080MB	36GB	600GB
RimU **	$67.95	1150MB	16GB	150GB
Media Temple (gs)	$80	1024MB***	100GB	1TB
Media Temple (dv)	$67.95	512MB	40GB	1.5TB
GoGrid ****	$149.98	512MB?	10GB	200GB

One of the challenges with making these comparisons is each of the services have slightly different offerings and different hardware configurations. For example GoGrid does offer Hardware Load Balancing, but many websites won't need this. Some of the services have more polished control panels and others appeal more to Linux geeks whole like full control. If we were to pick a winner today it would be Linode. They have several unique offerings such as disk images that allow custom partition schemes, configuration profiles to store muliple profiles and the Linode Shell that allows out of band access to your instance for emergency repairs. We are planning to change our hosting provider soon to Linode and will post a review shortly. Let us know what you think. * For Amazon we calculated the price as the base price per month (approx. $27/month) plus the hourly cost of keeping the server running 24/7 in a month. The calculation then is (30 days * 24 hours) * ($0.03 per hour) = $21.6. The base cost then is ($27.083 + $21.6) = $48.68. On top of that we calculated approx data transfer as Data In (50GB * .10) = $5 and Data Out at (100GB * .17) = $17 for a total data transfer cost of $22. ** For RimU we choose the customized option to add extra storage and transfer space *** Media Temple (gs) Ram was based on the advanced specs for GridContainers. **** For GoGrid we used the Business Cloud PrePaid Plan of $99.99 / month plus $49.99 / month for 200GB of outbound transfer

I have decided to stop development on anvil flex.  It had a good 2 year run, but it's sorely out of date.  Mostly it has been replaced by other frameworks.  Flex has made dramatic changes in the past several years and made a lot of anvil out of date.  The Spring BlazeDS project provides much better back-end support and integration.  On the Flex side the biggest downside is most of the data is passed using custom events.  Today I would rewrite this using something like Swiz. It has been really interesting developing anvil flex and I have learned a lot.  Probably the biggest lesson is how hard it is to keep an open source project up to date.  It gives you a much deeper appreciation for major open source projects and the amount of commitment it takes. Some parts of anvil flex are still useful.  It allows dynamically loading and managing modules.  It also has a unique portal framework.  The code will remain available on sourceforge and provides some good sample projects.