• A web front-end written using ASP.NET MVC 3 (and jQuery and jQuery UI) deployed on Windows Azure
• SQL Azure database
• Entity Framework
• Server processes written in C#
• Winforms-based administration application (using DevExpress components) with OData web services to the database
• Amazon Simple Queuing Service
• Python running on industrial controllers
• And tools and techniques: Resharper, Red Gate tools, dependency injection, xUnit and moq

C# skills are essential, but other than that the ability to quickly learn and use technologies is more important than past experience. You’ll be able to work either at their which are near us in South Lincolnshire (near Peterborough) or from home.

Tomorrow (Tuesday 12th July) I am in good old Birmingham NxtGenUG giving a talk on a Guerrilla Tactics approach to performances testing. Not how you would like your project to be but how to cope when they are not.

Below are the links to the slide, demos and other resources.
I used Apache Jmeter to do the multi threaded demos and there are link to that as well.

Have fun fighting those performance fires.

If you have any problem with the demos feel free to leave a comment, email or tweet. (I have to confess I am not great at using twitter these days.)

Links

• Guerrilla Warfare?

• Scripts for Guerrilla Warfare demos (with readme.txt)

• ClearTrace (free SQLServer Profile Analysis software)

• Trace Analyzer (free SQLServer Profile Analysis software)

• Brent Ozar-blog

• Apache Jmeter

• jTDS – SQL Server and Sybase JDBC driver (for Jmeter to connect to SQLServer)

The obvious answer, as always with these things, is “it depends”. But I was intrigued to see if I could quantify that a little further.

The batch profiling is done by setting up a server side Profiler trace. There is a post on how to do this here. Although ‘Export – Script Trace Definition’ from the SQLProfiler GUI does the job for you.

I have a little test rig here running SQLServer 2008 R2 and a load testing machine. I ran a simple load test creating orders. It had 20 threads and produces 4000 orders in just over 11 minutes.

I ran the whole test several times in three different scenarios:

• No tracing
• A GUI based Profiler trace logged data to screen and file on a remote machine
• A Server-side Profiler trace logging data to a local file on the server

The two traces logged the same 190,000+ events and captured the same columns.

My findings aren’t very surprising but here they are:

So the GUI profiling slowed the run down by 5.4% and the Server Profiler only by 2.8%.

The one thing that did surprise was the server profiler trace file, while holding all the same information, was 80% of the size of the file produced by the GUI.

The GUI has some advantages over Server Profiler:

• it is easier to use
• simpler to adjust the information being gathered
• traces can be written directly to a database table
• traces files are saved on the client PC, no server file system access required

Also in other cases the load the SQLProfiler put on the server would be low enough not to raise concerns.

However we can see that using server side profiling, logging to a local file on the server, can give you a real performance advantage over using the GUI.

It is quite a simple script the only parameter is @TotalCustomersPerSex. This can be set anywhere between 1 and 50,000 to give unique names. If you want some duplication or more name you can increase it or mess about with the source data.

The script has been written to target MS SQLServer 2008 although it could be altered to work with earlier versions of SQLServer.

The script does the following:

• Create a customer table
• Populates a temporary table with the 500 most popular surnames in England, Wales and the Isle of Man. (From the very helpful site http://surnames.behindthename.com.)
• Populates another temporary table with the 200 most popular boys names from England & Wales, randomly assigning a number I later use to populate the title.
• The same is then done for 200 girls names. The sources for both of these list was from http://www.statistics.gov.uk.
• I then create and populate a temporary table for males tiles and then repeat the process for females.
• I then have an insert statement selecting from forenames and titles and then cross joining the surnames table. This gives a Cartesian product.
• To populate information such as customer reference, initials, Post code, age, email address and password I extract characters from the forename and surname

I use the RAND function to populate the title field on insert is because this is only evaluated once per query. So if it is used in the final select only one random value would be selected. Not what I wanted.

I hope you find this useful.

Customer Generate script

Tonight (Tuesday 26th October) I am giving a talk on a Guerrilla Tactics approach to performances testing. Not how you would like your project to be but how to cope when they are not.

Below are the links to the slide, demos and other resources.
I used Apache Jmeter to do the multi threaded demos and there are link to that as well.

Have fun fighting those performance fires.

If you have any problem with the demos feel free to leave a comment, email or tweet.

Links

• Guerrilla Warfare?

• Scripts for Guerrilla Warfare demos (with readme.txt)

• ClearTrace (free SQLServer Profile Analysis software)

• Trace Analyzer (free SQLServer Profile Analysis software)

• Brent Ozar-blog

• Apache Jmeter

• jTDS – SQL Server and Sybase JDBC driver (for Jmeter to connect to SQLServer)

Last night (Monday 13th September) I did gave a talk on a Guerrilla Tactics approach to performances testing. Not how you would like your project to be but how to cope when they are not.

Below are the links to the slide, demos and other resources.
I used Apache Jmeter to do the multi threaded demos and there are link to that as well.

Have fun fighting those performance fires.

If you have any problem with the demos feel free to leave a comment, email or tweet.

Links

• A Grand Performance Slides

• Demo scripts (with readme.txt)

• Demo data (~5MB zipped)

• ClearTrace (free SQLServer Profile Analysis software)

• Brent Ozar-blog

• Apache Jmeter

• jTDS – SQL Server and Sybase JDBC driver (for Jmeter to connect to SQLServer)

There were two highlights for me:

• The critics of our own presentations
• The ‘How to Give Great Demos’ session

The critics of our own presentations, there was a good balance between not crushing our egos and giving us something to work on.

The ‘How to Give Great Demos’ session

I had to laugh at this because about the first Guy said was “don’t, whatever you do, use a laser pointer”. Earlier this year I was presenting at Microsoft Cambridge and it was suggested that I needed a laser pointer as the screen was huge and I am use to training rooms where I can point to the screen.

He had good reasons and now I have been told I’ll start playing with Zoomit, as suggested.

Other pearls of wisdom were:

• Introduce the context of the demo
• Demo code is not production code, just the code to make the point
• Don’t apologise that it is not production code and get on with the demo
• Show the finished product so they know where you are going (unless you want to surprise them)
• Keep the audience informed when you are typing
• Slow the mouse down so it is clear what steps you are taking
• Don’t use tools that aren’t part of the standard product e.g. Resharpen
• After the demo have a slide with the one thing you want to get across in the demo
• Record the demo as a fall back, perhaps you need an internet connection and that isn’t guaranteed
• Make code Lucida Console and set the size to 14 or 16
• Increase the DPI (maybe)
• Set the default highlight to Black text and yellow background
• Set the screen resolution to the project size you are going to use 2 weeks before the presentation
• Create a user on your laptop with all these settings
• Dave McMahon radically suggested developing all the demos on the laptop so you know they will fit

This and a lot more is available on Guy’s site.

http://www.guysmithferrier.com/downloads/HowToGiveGreatPresentations.pdf

This reminded me of a chat I had with a friend whose company database had just hit the 1 Terabyte size when portable 1 Terabyte USB discs came onto the market. He was the security manager there and was locking down USB devices in the same way.
The other day I bought a 1 Terabyte USB2 hard disk and I thought I would see how this, now ubiquitous, device stacked up against the old-school floppy.

How long to copy a Terabyte?

It took 9 hours to fill the disk. Not quick, but could easily be accomplished within a working day. But how long would it take to copy a Terabyte to floppy disk?

1 Terabyte is 1,000 x 1,000 x 1,000 x 1,000 = 1,000,000,000,000 bytes.

1 floppy disk can hold 1,457,664 bytes.

So to hold a Terabyte you would need 686,030 floppy disks. 16 floppy disks stack 5 cm (2 inches) tall. So your stack would be 214,384 cm high – over 2 kilometers (1.33 miles). That’s quite a stack of disks!

Carrying the Disks

People carry sports bags to work all the time, so wouldn’t look out of place. 75 litres is a standard size in the U.K., so let’s use one of those. Each disk is 8.89 x 9.40 x 0.32 cm = 26.74 cubic centimeters, so we can hold 37.4 disks per litre or 18345 litres to carry all our disks.

So a mere 245 trips using our sports bag will do the trick.

How Long To Copy The Data?

By experiment, I managed to fill 7 floppy disks in just under 6 minutes. So not allowing for any comfort breaks you should be able to fill 1 sports bag of disks in 6 hours 40 minutes. So if you take a sports bag of disks in to work per day, you’ll be able to walk out with a Terabyte of data in 49 weeks.

Of course, after spending nearly a year of spending your working days swapping floppy disks somebody will probably have noticed and rumbled you. Although after carrying over 46kg (100lbs) a day you may be able to hold your own with a security guard for a bit .

Just a bit of fun, please don’t try this at home. Or the office.

I am doing a Nugget on T4 at the NxtGenUG meeting at Coventry on 12th July 2010. The main event talk is Dave Sussman on How Clean is your ASP.NET?

The Nugget is a brief introduction to T4: what it is, what you can do with it and how it can make boiler plate coding far less dull and error prone. There have been a few minor changes since the nugget I gave in April. The demo project is VS2010 and uses a EF4 model rather than WCF data services model.

Links

• Demo code NxtGenT4Demo2010.zip

• Slides from the Coventry Nugget (pdf)

• T4 Editor

• T4 toolbox

• Oleg Sych

  T4 templates and articles

• T4 on MSDN

• Adventure Works 2008

• T4 to add INotifyPropertyChanged

  Use T4 code generation to add INotifyPropertyChanged to WCF Data Services (Astoria) Entities

I am doing a Nugget on T4 at the NxtGenUG meeting at Cambridge on 14th April. The main event is Jesse Liberty on Building A Highly Extensible, Decoupled Silverlight Open Source Application.

The Nugget is a brief introduction to T4: what it is, what you can do with it and how it can make boiler plate coding far less dull and error prone.

Links

• Demo code NxtGenUG-T4.zip

• Slides from the Nugget (pdf)

• T4 Editor

• T4 toolbox

• Oleg Sych

  T4 templates and articles

• T4 on MSDN

• Adventure Works 2008

• T4 to add INotifyPropertyChanged

  Use T4 code generation to add INotifyPropertyChanged to WCF Data Services (Astoria) Entities