The problem I encountered was that Win7 was not booting on this pc. This is because on the Win7, Microsoft changed the boot sector (don’t ask me why) and older systems do not recognize this as a valid bootable disk and the only message I was getting was a “cannot boot from cd code 5” error. After scouring the net I finally found a sollution which I want to post here.

1. Get a Windows Vista boot disk. It does not matter it’s from a “friend” or yours. You won’t actually intall Vista, just piggy-back on it’s boot capabilities.

2. Boot from the Vista disk.

3. After selecting the language select repair. Afterwards, depending on the state of your harddrive and other stuff you should be getting the choice of running a command prompt. Do it.

4. Change the disks such that your drive now contains the Windows7 disk.

5. In the command prompt navigate to the drive with the Windows7 disk, go to the sources directory and type setup.exe and press enter.

6. Now the Windows7 will start installing and everything will go on normally.

Happy installing!

 At that time I stuck with the Intel stock cooler which of course proved too loud. From the perfomance it was ok. I am not an overclocker so it was enough for the job. Still I needed to change it because it was loud as hell. I chose A Scythe Ninja 2 which can normally be used also in a fanless mode but I did not risk and installed also the provided fan, which is a 120×120 – 800rpm one (very quiet also).

My shock was of course that my system was as loud as before so I turned my head to the cooler on the VGA. This was a Saphire 4850 with a Zalman cooler; Noisy as hell as I saw, and thus I had to go for yet another upgrade which turned to be an Arctic Accelero S1 Rev.2. Also this is designed as a fanless sollution but again I wanted to play safe and added the turbo module which consists of 2 low rev fans barely audible.

All in all the system is now quiet and yet powerfull to play anything I throw at it. I could not make it dead silent but its miles from where it started. Below are some pictures I took along the process.

Scythe Ninja 2

Arctic Accelero S1 Rev.2 + Turbo module

Final view - don't mind the cables ... subject for new mod soon

Fortunately there is Microsoft Accelerator which you can get from here. This is a .Net abstraction over the GPU and allows for some basic computations to be forwarded to the GPU instead of the CPU. Naturally I wanted to explore a little bit and especially to do some tests. Below is my test example, some results and remarks.

using System;

using Microsoft.Research.DataParallelArrays;

namespace MSAccelerator

{

    class Program

    {

        public const int dimension = 5000;

        public const int repetitions = 1000000;

        static void Main(string[] args)

        {

            Microsoft.Research.DataParallelArrays.ParallelArrays.InitGPU();

            float[] matrix1 = new float[dimension];

            float[] matrix2 = new float[dimension];

            float[] matrix3 = new float[dimension];

            using (Timer t = new Timer())

            {

                Console.Out.WriteLine(“Starting generating the source matrices”);

                Random r = new Random((int)DateTime.Now.Ticks);

                for (int i = 0; i < dimension; i++)

                {

                    matrix1[i] = (float)(r.NextDouble() * 1000);

                    matrix2[i] = (float)(r.NextDouble() * 1000);

                }

                Console.Out.WriteLine(“Finished generating the source matrices.

                  Time taken (ms) :” + t.Milliseconds.ToString());

            }

            using (Timer t = new Timer())

            {

                Console.Out.WriteLine(“\nStarting adding the two the source matrices”);

                for (int k = 0; k < repetitions; k++)

                {

                    for (int i = 0; i < dimension; i++)

                    {

                        matrix3[i] = matrix1[i] + matrix2[i];

                    }

                }

                Console.Out.WriteLine(“Finished adding the two the source matrices.

                  Time taken (ms) :” + t.Milliseconds.ToString());

            }

            // transfer the data in structures than can be used by the accelerator

            DisposableFloatParallelArray pmatrix1 = new DisposableFloatParallelArray(matrix1);

            DisposableFloatParallelArray pmatrix2 = new DisposableFloatParallelArray(matrix2);

            DisposableFloatParallelArray pmatrix3 = new DisposableFloatParallelArray(matrix3);

            using (Timer t = new Timer())

            {

                Console.Out.WriteLine(“\nStarting adding the two the source

                  matrices using the GPU”);

                for (int k = 0; k < repetitions; k++)

                {

                    object result = ParallelArrays.Add(pmatrix1, pmatrix2);

                }

                Console.Out.WriteLine(“Finished adding the two the source

                  matrices using the GPU. Time taken (ms) :” + t.Milliseconds.ToString());

            }

            Console.In.Read();

        }

    }

    public class Timer : IDisposable

    {

        private DateTime startTime;

        public Timer() { startTime = DateTime.Now; }

        public int Milliseconds

        {

            get

            {

                TimeSpan ts = DateTime.Now.Subtract(startTime);

                return (int)(ts.Ticks / 10000);

            }

        }

        public void Dispose() { }

    }

}

The code above needs some explanations. The test involved the adding of two one dimensional int arrays. The dimension of the arrays were chosen to 5000 due to memory constraints. I am unable to pinpoint now the cause but going over 10.000 elements results in a null pointer exception in the Accelerator implementation. In might have something to do with the translation of the data into the vertices matrix or a bad implementation of the library. Because of this I chose to do a repetition in order to attain long running operations (1.000.000 repetitions). Yes this is quite high but at 10.000 repetitions the execution was under 15 msec and that is also around the value of the execution of the DateTime; so I had to choose a bigger repetition number.

In the bigger picture there are 4 steps: fill the matrix with random data, compute using a standard cpu-bound algorithm, put the data into structures usable by the Accelerator, make the operations using the Accelerator.

And now the results:

• on the cpu-bound algorithm: 36.909 msec
• on the gpu-bound algorithm: 265 msec

This results in a 139x increase in speed. Of course this is algorithm and hardware dependant.

Test machine specs: Core2Duo 6750 2.66Ghz, 4Gb Ram DDR2, Saphire ATI 4850 512Mb stock voltages/frequency.

One more thing to be added here are the obvious limitations of GPU processing, mainly the operations possible being on scalar values (int, float, double, bool). Also the context passing from CPU to GPU is costly and you only gain tabgible benefits only on arrays larger than 1 billion.

[UPDATE]: I found also a C# implementation of CUDA. I have not had time to look at it but it looks promising. Maybe in the future I will do some test comparissons between it and MS Accelerator.

Some days ago I read an interesting article about SSD performance degradation on Anandtech, and I kinda felt that this was happening to me also. I don’t own an SSD yet but I encountered flash performance degradation on my 8Gb Flash Voyager stick. Now I won’t go into the details about why flash performance degradation because the article above is more explanatory than I could make it. Suffice to say that I wanted to see if I can bring my stick back to life.

What was wrong with my voyager is that in the last weeks the write performance was poor. Whereas normally I could write at around 9MB/sec now I could only do it only at around 2-3MB/sec. From the article I understood that this was because even if the OS was reporting enough free space on the drive, all the space was already holding data and in order to write over it the flash needed to get the invalid data, write it to cache, delete the old location and then overwrite it with the new data. If one could delete everything from the location, subsequent writes would be just writes with no additional steps to execute, thus faster write speeds. This meant that I would need to do a format of the stick. But before that I had to do some tests to see performance before and after.

Let us see the scenario. Initially we have a full of data drive, even though the OS reported 7GB free out of 8Gb. I downloaded ATTO Disk Benchmark (I would have used HDTach but it does not work on my Windows7), and I ran a test of read/write for up to 256MB chunks. I stopped it when getting to 8Mb chunks as I saw no significant performance change going from 64K to 8Mb. You can see below the results:

Then I did a full format on the disk. Same FAT32 with 4096Kb allocation. The results on the write performance are in places 2.5 times greater. The read speed was the same. See below:

Of course format is not an action that you want to do often, as flash cells have a limited lifespan. But when a 2.5 times write performance penalty occurs I would certainly do it, especially when you really need this and paid top dollar for your flash (read Flash Voyager GT and the likes).

My testing machine specs (just so no one comes with remarks like “the test was bottlenecked somewhere”): Intel Core2Duo 6550 2.6Ghz, 4Gb DDR2 Corsair, Gigabyte P35 motherboard and a 500Gb Caviar Black HDD.

Cheers

• Analytics is Hot.  “The 21st century is all about math: some of the most unique, innovative ideas are emerging with these types of analytic projects. This is where the next billion dollar industries are being born.”
• Small is the new R&D.  “Today, the global R&D process has changed, and small is big. The global, infinite idea loop allows topic experts to share their latest research and insight with their global peers on a continuous basis. It’s a fundamental transformation in which most new scientific discoveries now percolate from the bottom up.”
• Attitude and Amusement. “The fact is, you’ll need them. That’s why workforce engagement is the big issue — you’ll only be able to get the staff you need if you can keep them active, engaged, interested and amused. A entirely different workplace concept that is radical, yet necessary.”
• Time Disappears.  “The major trend going forward is the collapse of time. There’s no time to plan anymore – there is just a need for action. While we still need budgets to manage and control, they’ll have to be constantly adjusted to deal with new realities. In this context, volatility is the new normal : the concept of risk management, for example, is transitioning quickly to one of risk containment.”
• Resistance to Change Retires.  “The coming generation of senior management aggressively pursues and implements new ideas. While the first is reluctant to embrace new business models, the next steamrollers them. Expect velocity!
• Careers End. “Your paycheck will come from: the global, itinerant, part-time, skills-for-hire economy.”
• Knowledge & Skills Banks.  “The capital of the 21st century isn’t financial : it’s experiential knowledge that is extremely scarce and specialized.”
• Interactivity Redefines.  “Every industry will soon be transformed by the forthcoming era of “pervasive connectivity.”  Essentially, every device and thing around us is about to become plugged in — leading us to an era of interactivity and connectivity that is mind-boggling in scope.”

Check this out : http://www.thewebsiteisdown.com/salesguy.html

Vista is the most stable Windows version I worked with and even though it’s a resource hog, on a high end machine it moves like it should.

Cheers.

I just read a small article on slashdot.org about Intel being on the verge to release 80 and 120Gb Flash drives (solid state drives). Beside this being a quite astonishing news, because we use to think Intel is making only processors and chipsets, one gas got me thinkinh of the current state of computing within the PC area.

I got now a rig with 4Gb Ram, and a SATA2 WD400Gb hardrive and still I don’t feel like my computer is flying or running extremly fast, although there are improvements. But this has a very solid reason – the harddrive. This is the slowest component in a system. The ram supports nowadays around 6400Mb/sec when using 800Mhz memory or 4267Mb/sec when using 533Mhz memory and the internal bus of a modern chipset can handle that easily; and dual core / quad core processor can process most available data on the bus. But all this hit a wall when communicating with the harddrive. Although the SATA2 interface can handle a theoretical 3000Mb/sec, and the best deliver around 2000Mb/sec, because of the physical constraints and how the drives are build, a typical read/write operation is around 1000Mb/sec and that is in ideal conditions.

So what can be done to improve this state of matters. You really have few options and all of them are expensive.

• Buy a 10.000 rpm drive. This is a simple sollution, but an expensive one as a 32Gb drive has the same price tag as a 500Gb 7.200Rpm drive.
• Get 2 smaller drives, let’s say 2x120Gb 7200rpm drives and put them in Raid 1 if your motherboard supports that. This can theoretically deliver around twice the bandwidth of a single drive both in writing and reading. This I think is the cheapest sollution at the moment because they can deliver more than a 10.000rpm drive in terms of bandwidth at a price twice as low.
• Get a SCSI harddrive. Well I am no expert in this but I know you get more performance when using it right but yes, it’s expensive. You have to have SCSI harddrives and also a SCSI controller which are not over the top in pricing but it will set you back quite some.
• Use solid state drives. Ugh, this is still a dream for the average user. SSD can deliver 0.1ms seek time and top models can sustain continuous 1000Mb/sec read operations while a normal mechanical drive has fluctuations of this value. Reading small files scattered around the disk is also faster that mechanical drives. Also there is small power consumption, better resistance to physical shocks, etc. Still there are downsides at the moment:
  • write operations are below the rate of a mechanical drive
  • price tag is simply prohibitive; a good SSD with 64Gb will set you back with 800-1000USD.
  • technology is still in it’s infancy. Many things can change and if manufacturers keep their promise, in two months you may realize your top of the line product is now rated middle segment.
• In very specific scenario the HDD replacement may be an overkill because there are better suited sollutions. In my case, a web developer, I saw cases where I could increase productivity a lot even if I could manage to move 1Gb of data from HDD to a more faster storage device. For example in RAM. Yes, why not? A good development machine has 4Gb of Ram. Windows, even Vista stays below the 1.5Gb mark so you have 2.5Gb of Ram that can be put to good use. So you can emulate a partition in Ram. There are tools for that and Microsoft even delivers a free one. Just move your projects there and set the database tempdb to be stored on that partition and voila, you have them on a 300 times faster drive.

So, there is a problem and there are sollutions. The choice depends really on the person, scenario and budget. At this point I would recoment 2 drives in Raid1 if you need extra performance but that’s just my pick, you have to find yours.

If you have better ideas, let us know …

Update

I just found a news on TomsHardware (here) about OCZ anouncing a SSD with SATA2 capable of sustained 120Mbps reads and 100Mbps writes which is almost twice as fast as the old SSD generation and it’s available already in the UK with prices starting around 700USD for the 32Gb model and 1300USD for the 64Gb model.

I banged my head for a couple of hours and for some reason I could not make it work. Lastly I decided to google for url rewrite AND IIS7. To my surprise there is an additional step to be done when registering your HttpModule (the one that will handle the url rewriting). You not only need to add it to the modules tag of the web.config, but also to the <system.webServer> modules tag. See example below (of course this is a stripped down code, you have to intertwine it with your existing web.config)

<system.web>
    <httpModules>
        <add name=“UrlRewriter“ type=“Rewrite, UrlRewriter, Version=1.0.0.0, Culture=neutral, PublicKeyToken=null“ />
    </httpModules>
</system.web>
<system.webServer>
    <validation validateIntegratedModeConfiguration=“false“ />
    <modules>
        <add name=“Rewriter“ type=“UrlRewriter.Rewrite“ preCondition=“managedHandler“ />
    </modules>
</system.webServer>

Lots of thanks to Scott Guthrie for his article on url rewriting in asp.net: http://weblogs.asp.net/scottgu/archive/2007/02/26/tip-trick-url-rewriting-with-asp-net.aspx

and also to Denis van der Stelt for his observations on IIS7 and url rewriting: http://bloggingabout.net/blogs/dennis/archive/2006/11/29/IIS7-and-Url-Rewriting.aspx 

07-03-2008
Yesterday I had to upload to a hoster a website using the above technique and it would not work. The hosting was using IIS6, so the ASP.Net would look at <system.web><httpModules> for additional modules, whereas IIS7 would not take into account that, but the settings in <system.webServer>. So what was wrong? The application complained that it could not load the type Rewrite. So I had a closer look and … I wrote the declaration wrong . Below you will find the correct add declaration to be added in httpModules:

<add name=“UrlRewriter“ type=“UrlRewriter.Rewrite, UrlRewriter, Version=1.0.0.0, Culture=neutral, PublicKeyToken=null“ />