second Tokyo Trip 2012 – Rakuten Technology Conference 2012

This October I had the pleasure to fly to Tokyo for the second time in 2012.

The development unit of Rakuten Japan was hosting the 7th Rakuten Technology Conference in Rakuten Tower 1 in Tokyo.

The schedule was packed with up to 6 tracks in parallel. From research to grass-roots-development a lot of interesting topics.

Source 1:
Source 2: Recorded Lectures

Mirror, Mirror on the wall

There are many things which are underestimated when team leads think about their team and possible actions to drive progress.

One of those things is that a team needs information to maintain and gain velocity. You cannot expect everyone to know just out of the blue what is important and in which direction everything is moving. To let everyone know and to develop that direction it’s important to share information as much as possible. It’s important to give everyone access to the information necessary to make a better job.

That’s why we had a build monitor at sones. We had a tool that displayed the current status of our build servers to all developers. Everytime someone committed a change, those build servers got this commit, built it and tested it with automated tests. The status of that could be seen by all developers as things happened.

So within seconds everyone could see if his commit did break something. Even better: Everyone could see. Everyone cared that the build needed to be working, that tests needed to pass. It was everyones job to do the housekeeping. When we switched from Team Foundation Server to GIT and Jenkins this status display needed to be replaced – you could immediately tell that things went from good to not-so-good in terms of build stability and automated testing.

Today I had the opportunity to take a tour of the Thomann logistics center. Standing in the support department I had this in front of me:

There were like 6 big status screens displaying incoming call status of the day, sales figures and other statistics important to those who work there. It’s a very important and integrated way to keep information flowing.

Since I am with Rakuten I thought about having a new status board set-up for my team. Something that might be inspired by the awesome status board which panic has built:

Since in addition to sones there are a lot of more things to track and handle (code, deployment, operations, overall numbers) I think such a status board will be of invaluable worth for the team.

Source 1:

Adventures in e-Commerce and technology

Oh dear. I just thought about the fact that I never really announced or talked about the fact that I changed my employee and moved to a (old) new place.

Yes that’s right, I am not with sones anymore. I am since January 1st the CTO of Rakuten Germany. When I signed the contract the company was called Tradoria – one of the first big projects I had the opportunity to work on was the so called brandchange.

A humongeous japanese based company called Rakuten bought Tradoria in the middle of 2011 and after half a year it was time to switch the brand.

As you can imagine these were busy weeks since January 1st. I had to digest a lot of existing technology and products. I met and got to know a lot of interesting people – first and foremost a great team of developers that went through almost all imagineable pains and parties to come up with a marketplace and shop system that is a perfect base for take-off.

A short word on the business-model of Rakuten – If you’re a merchant you gotta love it: Think of Rakuten as a full service provider for a merchant and customer. You as a Rakuten merchant get all the frontend and backend bliss to present and manage your products and orders. Rakuten takes care of all the nasty bits and pieces like hosting, development, telephone orders, invoicing, payment. The only thing that you as a Rakuten merchant need to do is to put in great products, gather orders and send out packages. Since Rakuten isn’t selling products on it’s own it won’t be competing with the merchants like other marketplace providers do these days.

On top of that Rakuten cares for the merchant and the customer. Just a week after that successful brandchange I attended (and spoke) at the Tradoria Live! 2012. That’s basically the merchant get-together. This year over 500 people attended this one-day conference. Think of it as a hands-on conference with features, plans, summaries of the last year and the upcoming one – every merchant is invited to come and talk to the people in person that work hard everyday to make the marketplace and shop system better.

click on it to see it big

Just 24 hours later standing on that stage I found myself here:


Yep. That’s Tokyo (東京). After a very long flight we had the chance to attend a all-embracing tokyo tour before the meetings and talks would start for our team. It was an awesome and exhausting week – just about 120 hours later I was back in Germany – I must have slept for two days :-)

Back in germany I had a lot of stuff to learn and work through. We had already moved to a wonderful house near Bamberg – it was pretty much big luck to find it. It’s actually ridiculously huge for a couple and two cats but we love it. Imagine the contrast: moving from an apartment next to a four-lane city street to the countryside just a 15 minute drive away from work with philosophical quietness all around.

Now after about half a year I am well into the process. I met a lot of high profile techies and things seem to take up speed in regards of teamplay in germany and with all the other countries. It’s a bliss to work for a group of companies that actually go through a lot of transitions while transforming from start-ups to an enterprise.

Ready for a family picture? Ready. Steady. Go!

That’s all Rakuten – that’s all on one mission: Shopping is entertainment! Empower the merchants!

Beside all that I even started to learn japanese. ただいま  :-)

benchmarking the sones GraphDB (on Mono (sgen) and .NET)

Since we’re at it – we not only took the new Mono garbage collector through it’s paces regarding linear scaling but we also made some interesting measurements when it comes to query performance on the two .NET platform alternatives.

The same data was used as in the last article about the Mono GC. It’s basically a set of 200.000 nodes which hold between 15 to 25 edges to instances of another type of nodes. One INSERT operation means that the starting node and all edges + connected nodes are inserted at once.

We did not use any bulk loading optimizations – we just fed the sones GraphDB with the INSERT queries. We tested on two platforms – on Windows x64 we used the Microsoft .NET Framework and on Linux x64 we used a current Mono 2.7 build which soon will be replaced by the 2.8 release.

After the import was done we started the benchmarking runs. Every run was given a specified time to complete it’s job. The number of queries that were executed within this time window was logged. Each run utilized 10 simultaneously querying clients. Each client executed randomly generated queries with pre-specified complexity.

The Import

Not surprisingly both platforms are almost head-to-head in average import times. While Mono starts way faster than .NET the .NET platform is faster at the end with a larger dataset. We also measured the ram consumption on each platform and it turns out that while Mono takes 17 kbyte per complex insert operation on average the Microsoft .NET Framework only seems to take 11 kbyte per complex insert operation.

The Benchmark

Let the charts speak for themselves first:


click to enlarge

click on the picture to enlarge

click on the picture to enlarge

As you can see on both platforms the sones GraphDB is able to work through more than 2.000 queries per second on average. For the longest running benchmark (1800 seconds) with all the data imported .NET allows us to answer 2.339 queries per second while Mono allows us to answer 1.980 queries per second.

The Conclusion

With the new generational garbage collector Mono surely made a great leap forward. It’s impressive to see the progress the Mono team was able to make in the last months regarding performance and memory consumption. We’re already considering Mono an important part of our platform strategy – this new garbage collector and benchmark results are showing us that it’s the right thing to do!

UPDATE: There was a mishap in the “import objects per second” row of the above table.

taking the new and shiny Mono Simple Generational Garbage Collector ( mono-sgen ) for a walk…

“Mono is a software platform designed to allow developers to easily create cross platform applications. It is an open source implementation of Microsoft’s .Net Framework based on the ECMA standards for C# and the Common Language Runtime. We feel that by embracing a successful, standardized software platform, we can lower the barriers to producing great applications for Linux.” (Source)

In other words: Mono is the platform which is needed to run the sones GraphDB on any operating system different from Windows. It included the so called “Mono Runtime” which basically is the place where the sones GraphDB “lives” to do it’s work.

Being a runtime is not an easy task. In fact it’s abilities and algorithms take a deep impact on the performance of the application that runs on top of it. When it comes to all things related to memory management the garbage collector is one of the most important parts of the runtime:

“In computer science, garbage collection (GC) is a form of automatic memory management. It is a special case of resource management, in which the limited resource being managed is memory. The garbage collector, or just collector, attempts to reclaim garbage, or memory occupied by objects that are no longer in use by the program. Garbage collection was invented by John McCarthy around 1959 to solve problems in Lisp.” (Source)

The Mono runtime has always used a simple garbage collector implementation called “Boehm-Demers-Weiser conservative garbage collector”. This implementation is mainly known for its simplicity. But as more and more data intensive applications, like the sones GraphDB, started to appear this type of garbage collector wasn’t quite up to the job.

So the Mono team started the development on a Simple Generational Garbage collector whose properties are:

  • Two generations.
  • Mostly precise scanning (stacks and registers are scanned conservatively).
  • Copying minor collector.
  • Two major collectors: Copying and Mark&Sweep.
  • Per-thread fragments for fast per-thread allocation.
  • Uses write barriers to minimize the work done on minor collections.

To fully understand what this new garbage collector does you most probably need to read this and take a look inside the mono s-gen garbage collector code.

So what we did was taking the old and the new garbage collector and our GraphDB and let them iterate through an automated test which basically runs 200.000 insert queries which result in more than 3.4 million edges between more than 120.000 objects. The results were impressive when we compared the old mono garbage collector to the new mono-sgen garbage collector.

When we plotted a basic graph of the measurements we got that:



On the x-axis it’s the number of inserts and on the y-axis it’s the time it takes to answer one query. So it’s a great measurement to see how big actually the impact of the garbage collector is on a complex application like the sones GraphDB.

The red curve is the old Boehm-Demers-Weiser conservative garbage collector built into current stable versions of mono. The blue curve is the new SGEN garbage collector which can be used by invoking Mono using the “mono-sgen” command instead of the “mono” command. Since mono-sgen is not included in any stable build yet it’s necessary to build mono from source. We documented how to do that here.

So what are we actually seeing in the chart? We can see that mono-sgen draws a fairly linear line in comparison to the old mono garbage collector. It’s easy to tell why the blue curve is rising – it’s because the number of objects is growing with each millisecond. The blue line is just what we are expecting from a hard working garbage collector. To our surprise the old garbage collector seems to have problems to cope with the number of objects over time. It spikes several times and in the end it even gets worse by spiking all over the place. That’s what we don’t want to see happening anywhere.

The conclusion is that if you are running something that does more than printing out “Hello World” on Mono you surely want to take a look at the new mono-sgen garbage collector. If you’re planning to run the sones GraphDB on Mono we highly recommend to use mono-sgen.

How To strip those TFS Source Control references from Visual Studio Solutions

Every once in a while you download some code and fire up your Visual Studio and find out that this particular solution was once associated to a team foundation server you don’t know or have a login to. Like when you download source code from CodePlex and you get this “Please type in your username+password for this CodePlex Team Foundation Server”.

Or maybe you’re working on your companies team foundation server and you want to put some code out in the public. You surely want to get rid of these Team Foundation Server bindings.

There’s a fairly complicated way in Visual Studio to do this but since I was able to produce unforseen side effects I do not recommend it.

So what I did was looking into those files a Visual Studio Solution and Project consists of. And I found that there are really just a few files that hold those association information. As you can see in the picture below there are several files side by side to the .sln and .csproj files – like that .vssscc and .vspscc file. Even inside the .csproj and .sln file there are hints that lead to the team foundation server – so obviously besides removing some files a tool would have to edit some files to remove the tfs association.


So I wrote such a tool and I am going release it’s source code just beneath this article. Have fun with it. It compiles with Visual Studio and even Mono Xbuild – actually I wrote it with Monodevelop on Linux ;) Multi-platform galore! Who would have thought of that in the founding days of the .NET platform?

Bildschirmfoto-StripTeamFoundationServerInformation - Main.cs - MonoDevelop

So this is easy – this small tool runs on command line and takes one parameter. This parameter is the path to a folder you want to traverse and remove all team foundation server associations in. So normally I take a check-out folder and run the tool on that folder and all its subfolders to remove all associations.

So if you want to have this cool tool you just have to click here: Sourcecode Download

Developing on a Microsoft Surface Table

At sones I am involved in a project that works with a piece of hardware I wanted to work with for about 3 years now: the Microsoft Surface Table.

I was able to play with some tables every now and then but I never had a “business case” which contained a Surface. Now that case just came to us: sones is at the CeBIT fair this year – we were invited by Microsoft Germany to join them and present our cool technology along with theirs.

Since we already had a graph visualisation tool the idea was to bring that tool to Surface and use the platform specific touch controls and gestures.

the VisualGraph application that gave the initial idea

The good news was that it’s easier than thought to develop an application for Surface and all parties are highly committed to the project. The bad news is that we were short on time right from the start: less than 10 days from concept to live presentation isn’t the definition of “comfortable time schedule”. And since we’re currently in the process of development it’s a continueing race.

Thankfully Microsoft is committed to a degree they even made it possible to have two great Surface and WPF ninjas who enable is to get up to speed with the project (thanks to Frank Fischer, Andrea Kohlbauer-Hug, Rainer Nasch and Denis Bauer, you guys rock!).

a Surface simulator

I was able to convice UID to jump in and contribute their designing and user interface knowledge to our little project (thanks to Franz Koller and Cristian Acevedo).

During the process of development I made some pictures which will be used here and there promoting the demonstration. To give you an idea of the progress we made here’s a before and after picture:

We started with a simple port of VisualGraph to the surface table…

…and had something better working and looking at the end of that day.

I think everyone did a great job so far and will continue to do so – a lot work to be done till CeBIT! :-)

Source 1:
Source 2:
Source 3: