Background information

First of all, the way http.sys works on newer Windows versions is a very useful feature. The reason behind http.sys is to have a single system service that listens on the all-important port 80 and be a mediator to different applications and services in the system that might like to handle incoming requests on port 80 but for different urls. What an integrated Windows application does, is use a Windows API for registering its own urls and then have them directed to itself. At the same time, there can be other applications also getting incoming requests to port 80 but for different registered urls.

The problem

The problem is that some (many?) open source project that have been compiled for Windows, does not use this http.sys API. Therefore, when they start up and try to bind to port 80, they will fail since it is occupied by http.sys. And I have to admit that I have no idea how easy or hard it would be to patch such a project to make that call on Windows and get the tcp/ip network stack to work with it.

Examples of such projects are:

• Apache HTTP Server
• Nginx

(Haven’t looked at how nodejs does it.)

The solution

Some solutions I’ve found suggests disabling http.sys, but that only works if you are sure you won’t have any use for Windows applications that do know about http.sys. I didn’t want to do that.

My suggested solution is instead to tell http.sys to bind to a specific ip-address, and then use any other ip-addresses on your system for Nginx (or whatever you are looking to install). It turns out this is possible via the netsh command line configuration tool in Windows (from version 7 anyway).

So, one way would be to add an extra ip address to your network configuration. But I chose instead to use the fact that every system nowadays has both an ipv4 and an ipv6 address. My solution is to bind http.sys to the ipv6 general address (named “::”) and thereby free up the local general ipv4 address (named “127.0.0.1″) for my Nginx server.

Telling http.sys to bind to the ipv6 general address is easy. Just run the follow command from a command prompt window as Administrator:

netsh http add iplisten ipaddress=::

It seems that the name “localhost” binds to the ipv6 address, actually “::1″, so from now on you can still get the http.sys applications and services via “http://localhost”. And if you install and run Nginx on port 80 it should start fine and be available at “http://127.0.0.1″. Or you could even edit your “C:\Windows\System32\drivers\etc\hosts” file and add the line:

127.0.0.1 local

Now having “localhost” for ipv6/http.sys and “local” for ipv4/Nginx.

My background is primarily in Microsoft.NET and Microsoft Visual Studio, so it did feel a little different when I first entered the world of Java and Android development. But once you get past those initial hurdles, it isn’t as different as you first might have thought.

These are the steps to get started with Android development on Windows. Should work on any Windows version, but I’m using Windows 7 64-bit.

1. Download and install Java SDK.
2. Download and unzip Android SDK.
3. Run “SDK Manager” from Android SDK to download the platform versions.
4. Download and unzip Eclipse.
5. Install the Android plugin for Eclipse and point to the Android SDK.

You can read the instructions here, or watch a screencast I made describing the same procedure:

http://vimeo.com/16392228

Download and install Java SDK

You’ll find the Java SDK here:

http://www.oracle.com/technetwork/java/javase/downloads/index.html

Click the “Download JDK” button and choose the Windows version you have from the Platform combobox. You don’t have to register to download. I ended up downloading a file called “jdk-6u22-windows-x64.exe”. Run it and accept all defaults to complete the installation.

Download and unzip Android SDK

The Android SDK doesn’t have an install executable, but rather a zip file and a utility called “SDK Manager.exe”. Begin by downloading the sdk from here:

http://developer.android.com/sdk/index.html

Unzip the file to any location you want. I put it in “C:\Android\SDK”. Inside that folder, you’ll find the executable “SDK Manager.exe”. Run it. Every time you run that application, it will look for updates to anything you’ve downloaded previously with the same tool. The first time it will default to download all platform versions of Android except the Google specific ones. I’d recommend selecting Cancel at this initial update dialog, and then download what you need from “Available packages” instead.

The Android platform has been released in several versions since its initial offering of version 1.0. Every Android application will require a specific version, but will of course work on all subsequent versions too (at least in a perfect world). Today, I recommend that you target version 2.1 as the least common denominator. But note that there might still be users and phones of at least version 1.6 that might be of interest to you. Anyway, this means that I would download version 2.1 and the latest version 2.2. They don’t really take up that much disk space (maybe ~100MB per version?), so it won’t hurt you too much to download them all either.

Also, every Android platform version comes in a “Android plain vanilla” variant and a “Google APIs” variant. The only differences between those two are that “Google APIs” variant will include the ability to use Google Maps component and some other Google specific APIs that some Android devices might not support. I found this naming to be a little strange at first, but note that “Google APIs” includes everything in “Android plain vanilla” too.

The point is that if your application doesn’t need things such as Google Map component, you will be targeting more devices if you choose to develop for “Android plain vanilla”. As far as I know, all commonly sold Android phones support the Google APIs and it is only simpler media players and maybe cheap tablets that might not have chosen to support Google APIs (since I assume the manufacturer pays licensing fees to Google for that).

Run “SDK Manager” from Android SDK to download the platform versions

So, running “SDK Manager”, cancelling on the first update dialog box, you’ll move on to “Available packages”. I recommend checking the following items and then click “Install selected”: (revision numbers can of course have changed since I wrote this)

• SDK Platform Android 2.2, API 8, revision 2
• SDK Platform Android 2.1, API 7, revision 2
• Samples for SDK API 8, revision 1
• Google APIs by Google Inc., Android API 8, revision 2
• Google APIs by Google Inc., Android API 7, revision 1
• Usb Driver package, revision 3
• Market Licensing package, revision 1

You’ll need to “Accept all” licenses and then the download should start. You might get a question if it is ok to restart ADB at the end of the installation procedure, which it is!

Download and unzip Eclipse

You don’t have to use Eclipse. You could stop here and use a command line utility from the Android SDK to create project skeletons and edit the files with any text file editor. But I think you’ll enjoy the full development environment of Eclipse, even though it seems to have gotten its share of complaints over the years.

Download Eclipse from here:

http://www.eclipse.org/downloads/

The first option of “Eclipse IDE for Java Developers” is what you want, and click the link to the right depending on what Windows version you’re on. I ended up downloading “eclipse-java-helios-SR1-win32-x86_64.zip”.

Eclipse doesn’t have an install at all. Just unzip the files into any folder you like and start the environment with “eclipse.exe”.

Install the Android plugin for Eclipse and point to the Android SDK

Inside Eclipse, you need to do a few things before you can create your first Android project. When you start Eclipse, you choose your “Workspace”. Just accept the default. This is the base folder where you will put all of your projects, and can of course be changed later on.

Now, do the following:

1. Select Help menu
2. Install new software
3. Add
4. Type any Name, like “Android”
5. Paste this address in Location: https://dl-ssl.google.com/android/eclipse/
6. OK
7. After a few seconds, the list should contain “Developer Tools” and there you’ll find “Android DDMS” and “Android Development Tools”.
8. Select both and then Next
9. Continue with the download and install. You’ll be asked to restart Eclipse at the end. Do that.

Back in Eclipse with the Android plugin now installed, you also need to:

1. Select Window menu
2. Preferences
3. Android
4. Click Browse button next to “SDK Location”
5. Point to the folder where you unzipped the Android SDK previously. For me, that would be “C:\Android\SDK” (where the “SDK Manager.exe” is located)
6. OK
7. Apply
8. After a few seconds the list should fill with the Android platforms you chose to download with “SDK Manager” previously
9. OK

That’s it!

Create your first Android project

Now you should be able to create your first Android project. Still in Eclipse:

1. Select File menu
2. New
3. Project
4. Android, Android Project
5. Next
6. Project name: My first project
7. Build Target, Target Name: check any platform version you’d like to start with
8. Application name: MyFirstProject
9. Package name: com.myname.MyFirstProject
10. Create Activity: (checked) MainActivity
11. Min SDK Version: (can be left empty)
12. Next
13. Finish (no test project for your first attempt)

Now, right-click the project name in the left pane and choose “Run as” and then “1 Android Application”. The emulator should start and after some time (it IS really slow to start), you’ll be running your first Android application. If the emulator starts, but not your application, just keep the emulator running and switch back to Eclipse and execute the “Run as” command again. Sometimes it fails on the first attempt.

I first wanted to get a better understanding of what a virtual machine looks like in Windows Azure. What is it exactly? I know it’s a Microsoft Windows Server box running on Hyper-V, at least that is what Microsoft has told me. But then what? Read on to discover some of my findings. Note, however, that I’m not an expert and I’m sure that some of my findings and questions could be answered by reading documentation somewhere but I like to get my hands dirty by looking under the hood.

Operating System Environment

My virtual machine (vm) in Windows Azure is running the following version of Windows::

Microsoft Windows NT 6.0.6002 Service Pack 2

Which translate to the marketing name of “Windows Server 2008″, that is NOT R2 (yet, anyway).

Also, it is running:

Microsoft.NET version 2.0.50727.4016

I have a Small VM and therefore should have 1,7 Gb of memory in total, which I also verified and got the exact number back:

1,877,766,144 bytes of total memory

Not that it matters but I also found my machine name to be “RD00155D3141A4″ and part of the domain “CIS”.

File System Organization

The organization of the file system is more interesting. On a newly created vm and deployed web role I can see three drives:

C: has a total of 225 GB of which 218 GB are free
D: has a total of 16 GB of which 8 GB are free
E: has a total of 1 GB of which almost all are free

After investigating their contents I can determine that D: has the installed operating system with the usual “Program Files” and “Windows” folders. E: has my deployed web role in the E:\approot folder and just some additional support/config folders and files (I assume!).

The C: drive is the stranger drive. It has a “dumpfile.dmp” and a “pagefile.sys” file which I recognize as Windows files. There are also folders with names such as “Applications”, “Config”, “MOSLogs”, “OS”, “Packages”, “Resources” and the known “System Volume Information”. Some are empty but not all (I’ll get back to that.

When I upgrade my web role with some changed code, the E: drive is kept an F: drive appears. So it looks like at least a few versions of deployed roles will be kept at different drive letters.

The different vm sizes specifies that I should have 250 Gb of storage for “Disk Space for Local Storage Resources” so this must be the C drive. I don’t have write access to all of it though, but I probably just haven’t found where you are supposed to put such files. In any event, I assume that you must be prepared for that storage to vanish at any given moment. For sure if you delete and recreate your vm.

Processes

I have also tried to enumerate all running processes in the machine and I ended up with the following list:

Not that interesting perhaps, but I can determine that my deployed web role has spawn the WaWebHost.exe and the MonAgentHost.exe processes. Maybe more information could be read, but I’m leaving this part of the investigation as it is for now.

Web Role

So how is the code in my web role actually run by Azure? I can see these things:

• The exact code for the built/published web project is stored in E:\approot. There are my Global.asax, Web.config, bin-folder and so on.
• The process that has been spawn to run this code is stored in E:\base\x64\WaWebHost.exe. I think this exe file is actually contained in the deployed packade, which means I can control this executable if I so wish?
• There is no w3wp.exe process running so it seems that IIS is not doing any work at all to run my code. I’m guessing though that WaWebHost.exe may be a special version/compilation of w3wp.exe for Azure?

Now, the most important question I would like to answer in this first investigation is if, and if so how, I can run multiple sites or web projects in a single vm. Knowing IIS, I should go looking for applicationHost.config and see how it points to the web site folders (E:\approot according to the investigation so far).

In fact, I find IIS in its usual place at D:\Windows\System32\inetsrv in my vm. There is also an applicationHost.config in the config sub folder, but it has nothing pointing to E:\approot.

However, I did find the file at C:\Resources\Temp\63a1a1bad2fd41729fe654d54ada5d05.PublicRole (a generated folder name that will look different on other vm:s, for sure). There I can see:

• An Application Pool that is running as “NetworkService” (the only one).
• A web site with the name “Role Site” and a mapping of the root “/” to “E:\approot\”.
• A binding for this site to ip 10.115.141.65 on port 20000 (no host name).

So my vm has a local network address behind Azures load balancer (or gateway or whatever it is called) and it is routing incoming traffic on port 80 to port 20000 on my vm.

The obvious question is if I can change this applicationHost.config file and thereby point it to sub folders in my web role, which would then actually represent different sites or web projects in Visual Studio. I just had to try! Unfortunately nothing happens when I added another site to this file. And since it is in a Temp sub folder I assume it is only used when WaWebHost.exe starts up. I also tried suspending the vm and than starting it again (not sure if this is the same as rebooting the vm?). But then the file is overwritten and my new site info is gone. Too bad, but of course if it was supported there would be an easier way to do it than such a hack.

Conclusion about suitable scenario

One thing I’ve come to realize is that Windows Azure (of today, at least) is not a substitute for a VPS or dedicated server of my own. My first attempt was to find out if Windows Azure actually could be the solution to all of my hosting needs, and it is probably not. Even if it would work, it is obviously not designed for handling multiple sites.

Windows Azure is for when one of your ideas or projects has gotten enough attention or quality that it is time to launch to the public and be prepared for both success and failure. You will be able to scale up to any size you need, and you will also be able to shut it down literally within minutes without any additional costs. Of course, you’ll have to pay for the time between those two events: the launch and the decision.

I will keep on investigating this and other matters around Windows Azure. I also have hopes that multiple roles or multiple sites/projects with different host name bindings in a single vm will be possible in the future.

So I started to investigate my cabling and restarting switch and router, but nothing helped. I then verified that my network was ok by starting my backup laptop and it did get an internet connection so everything was fine there.

[Skip to end if you just want the solution that worked for me.]

Next suspicion was that my network interface card in my Dell computer had died. It is a “Intel 82567LF-2 Gigabit Network Connection”. However, the signs wasn’t there that it might have died. Nothing wrong with it in Device Manager (Right click “Computer” and choose “Manage”).

Windows did recognize the difference between disconnected cable (red cross for network icon in task bar notification area) and my currently unsolved problem (yellow exclamation point) and “Unidentified network”. Also, the network cable lights at the connector seemed to work fine (at least they was lit). Still no network and no pings working. Trying to ping the router or another computer in my network timed out or gave “Destination host unreachable” from the local network driver.

The network had died just when I started Spotify, so I was now starting to suspect that it was something wrong in Windows. I have Windows 7, 64-bit. Some sort of configuration that had gotten the hiccups? I found some commands (“netsh”) that can reset the Windows tcp/ip info, but it did not help (after reboot). I removed the network interface card drivers in Driver Manager and restarted, at which point the drivers were reinstalled automatically, but the problem remained.

I also read that something similar had happed to someone and that he had reinstalled all of Windows and thereby gotten it to work again. Not something you really would want to do.

As a side note, my HTC Hero proved useful since I managed to copy the Android Phone drivers from my laptop to the desktop computer and thereby I could get online using my shared connection facility of HTC Hero and its usb cable. Nice!

Anyway, it struck me that network interface cards actually flash their lights when the power of the computer is off. So maybe it is the firmware in the Intel card that had gotten the hiccups?

Solution

So I shut down my computer and removed the power cable for a few minutes. Then I started it again, and what do you know:

It worked again!

Update: I finally got this error a handful of times and have now installed the following from Intel (not from Dell, since that download said that I already had the latest drivers). We’ll see if problem goes away completely or not. After this install, the driver version went from 10.0.0.2 to 10.1.6.0.

Intel drivers

• Download and install Java Development Kit SE: JDK 6 Update 16.
  File name I ended up downloading was jdk-6u16-windows-x64.
• Download and unpack Eclipse IDE for Java Developers.
  File name I ended up downloading was eclipse-SDK-3.5.1-win32-x86_64.zip.
• Download and unpack Android SDK.
  File name I ended up downloading was android-sdk_r3-windows.zip.
• Changed android.bat inside Android SDK according to this tip.
• Start a command prompt as administrator and run the following command from the tools folder of the Android SDK:
  android update sdk
• Fixed error “Failed to fetch URL https://dl-ssl.google.com/android/repository/repository.xml, reason: HTTPS SSL error. You might want to force download through HTTP in the settings.” by doing as it says and then clicking Update all button under Installed packages. Not sure why this was needed because the https link loaded fine in my browser.
• Select “SDK Platform Android 1.5″, Accept and then Install Accepted. Actually I ended up downloading all platforms (not sure why, but I have room so what the heck). This means 1.1, 1.5, 1.6, 2.0, documentation
• Start Eclipse and accept the default workspace location (or change – it is just the place to put your projects and files). Choose Help, Install new software. Input the following link and click Add button.

  https://dl-ssl.google.com/android/eclipse/

  Type a name for the link/source and ok. Check “Developer Tools” once it appears in the list and click Next, Next, I Accept and Finish. I got a warning about unsigned content, but accepted with OK. Also, a Yes for restarting Eclipse.

• Choose Window, Preferences inside Eclipse. Under Android, Browse for the location where you unpacked the Android SDK. I clicked Apply, but it seemed to take a while and maybe another click and/or another Browse attempt before the list filled up with my targets “Android 1.1″, “Android 1.5″ etc. OK.
• Choose Window, Android SDK and AVD Manager inside Eclipse. Click New under Virtual Devices. Type a name for your phone emulator and select your target platform, Android 1.5 for me. Type a size for your emulated SD card, I chose 200 MiB. I did not create any additional hardware abstractions (yet, will probably come back later). Then Create AVD and close the dialog box.
• Choose File, New, Project inside Eclipse. Select Android Project and Next. Type a name for your project and select your target platform. I’m going for Android 1.5 which is on my HTC Hero at the time of this writing. Also input values for the following fields, for example:
  Application name: My App
  Package name: com.mycompany.myapp
  Create Activity: MainActivity
  And complete with a click on Finish. Now, you may think nothing has happed, but it has. Just click icon on right which saids “Workbench” when you hover on it. There you have your project and source files to work with!
• Click Debug button in toolbar in order to test the application in the emulator. Select Android Application and off you go. Even on my Intel Core i7 processor it takes some time to get the app running inside the emulator, about 1 minute! Note however, that if you keep the emulator running, it only takes about 5 seconds to launch a new version of my application from within Eclipse.

That’s it. Now off for some educational reading in how to write Android applications. I hope to get back on that matter.

I don’t think it is relevant for this setup, but I have downloaded and installed the latest update for HTC Hero which is ROM Upgrade for HTC Hero_2.73.405.5.

I have also installed the HTC Sync software on my Windows 7 laptop (update: now also working on my desktop Windows 7 64-bit machine).

My current 3G subscription is from Telenor (Sweden).

Everything works fine on my phone when connecting over the 3G network.

So, in order to set this up I connect the usb cable to the laptop and then I turn on the following checkbox setting in Android:

• Settings
• Wireless controls
• Mobile network sharing

Windows immediately recognizes a new usb device and starts to install it but fails with “No drivers for device Android Phone”. What a bummer!

Trying to find an updated driver automatically also failed. So I started searching for driver files but soon realized that there is actually a folder on disk after the HTC Sync installation that is called “Driver Files”. It is located here on my machine:

C:\Program Files\HTC\HTC Driver\Driver Files\Vista_x86

So I chose to manually browse for drivers instead, and directed the Windows dialog to this path and what do you think? It worked!

Update: if you can’t find the mentioned files, I’ve zipped them together here (both 32-bit and 64-bit versions, should work on both Vista and Windows 7).

Simple, but yet a useful tip.

“Hello” – someone says. What? Hey – guess what – it’s the customer!

“You’re early!” – I said.

“I am?, I thought we said eleven o’clock” – he replies.

But I’m not finished and the clock in my right hand corner saids “10:24″. Aaargh. What’s going on? So I click on the damn clock and guess what, the time is REALLY “11:00″. Great!

Now, this has actually happend once for me and once for my coworker. Nothing major of course, but it is really annoying that you can’t trust the Windows Vista clock anymore. Either you have to click on the time in the right hard corner to make sure that it is really, really, the correct time – or maybe invest in a wrist-watch…

An error occurred on the server when processing the URL. Please contact the system administrator.

Hey, that’s me! And then:

Microsoft JET Database Engine error ’80004005′
Unspecified error
/myapp/myscript.asp, line 9

Ok, so I suspected that the first problem was due to the fact that I needed to activate asp scripts in some way. However, turns out it was a new setting for sending the error message to the client (maybe it has existed somewhere before, but I’ve never had to change it). This setting is located in the new IIS Manager:

Features View, ASP, Debugging Properties, Send Errors to Browser

But this only got me to the second error message. And that, I had seen before. It should have something to do with file and directory permissions. Being uncertain about which user IIS 7 actually runs legacy asp scripts as, I used Process Monitor to find the “IUSR” user. But givning that user full permissons to the directory where my.mdb resided didn’t help. So back to Process Monitor – filtering on failed operations from the w3wp.exe process. Ahh! Temporary files. Of course! The OLEDB driver needs access to the system temporary folder to store some files when opening the connection to the database file. And the location for this in Vista is:

C:\Windows\ServiceProfiles\NetworkService\AppData\Local\Temp

Now, UAC can really drive you mad. I wonder how long I can take it before giving up and turning it off. For security purposes, it is of course great. Then again, I’ve run Windows for a lot of years as a member of the Administrators group, and never had any real security issues. At least not any that I know of…

Pasting “C:\Windows\ServiceProfiles\NetworkService\AppData\Local\Temp” into the address bar of Explorer will tell you that it doesn’t exist. Well, I didn’t believe Vista so I removed the ending directories until Vista told me that I had to elevate priviliges to get into one of the parent folders. Ok, I did that and now I could get into the full path. Great. A warning that I was changing file permissons on a system directory was all that was standing between me and success – and thankfully that was not an obstacle.

So now I’m running my old asp-scripts until I’ve had the opportunity to convert all projects to asp.net.