.NET/Microsoft detractors got an early Christmas present recently when the London Stock Exchange, under a relatively new CEO, decided to dump their .NET/SQL Server –based trading platform, TradElect, to replace it with the product of a being-acquired company.

On Slashdot, news of this was submitted and accepted as “London Stock Exchange Rejects .NET for Open Source”, with the statement that “The switch is a pretty savage indictment of the costs of a complex .NET system.” The Digg submitter went with the title “London Stock Exchange dumps Windows for Linux” — which they took directly from the linked article — with the description “Fed up with Windows' failures, one of world's major stock exchanges is joining many others in making the switch to reliable Linux“.

The heavily-linked columnist in both cases is a guy who has been riding this "LSE dumps Windows!" horse for a while now. It has certainly provided him with lots of quality incoming visitors, drawing in those looking for validation, and angry hordes baited by his trolls. Encouraged on he seems to be accelerating the unsubstantiated hyperbole.

Let’s take a moment to go back in history for a bit.

Microsoft made a Really Big Deal about the LSE originally switching to this custom, Accenture-built, SQL Server 2000/.NET-based solution. This was sort of Microsoft’s coming out party, in a way saying “look, we’re big boys too! No more pull-ups for us”.

When the LSE had a very public failure on one of the biggest trading days in history, the detractors were screaming “I told you this would happen!” until their throats were sore, despite the cause of the failure never having been publicly detailed.

Failures have happened on every platform, most commonly as a result of application failure. To automatically assume the worst of a novel solution simply because it is atypical is the thought process of annoying simpletons, anxiously and eagerly hoping to try to pin any fault on anything that doesn’t fit their vanilla perspective of how things are supposed to work.

“The software must have failed because Bob went with HP instead of IBM!”

“The payroll system miscalculated. It must be because they moved it to Linux from Solaris!”

So what really happened with the LSE?

Accenture built a very expensive, custom solution for the LSE, purportedly costing somewhere in the neighborhood of $65 million dollars. To operate this custom in-house (albeit designed by Accenture out-of-house) system the LSE built up a considerable technology workforce.

The worldwide recession hits and the LSE takes some financial hits. A swarming mass of competitors in Europe, many running off-the-shelf, superior systems that they’re paying less for, go live.

A new CEO takes over and immediately starts to swing the axe. He makes specific comments in the press about the high IT costs of the organization; both of the large number of technology workers in London, and the continuing significant payments to Accenture to finesse the TradElect platform.

He undoubtedly observed that all of this custom work hasn’t gained them any unique advantage in the relatively commodity task that they performed. In some ways it’s like writing and maintaining your own in-house operating system – if it doesn’t give you some advantage, and actually puts you behind as everyone else pools resources on a solution, then why would you do that?

So they go on the market for a replacement, eventually deciding to go with the product of a Sri Lankan company. The price is right, and the lure of low-priced Sri Lankan talent is enticing enough that they buy the whole company.

In the end they have switched from an extravagant, custom-developed solution built by a notoriously expensive consulting company, and a workforce of expensive talent in the West, to a basically off-the-shelf solution that has been subsidized to its current state by other organizations, in the process getting some low-paid talent in South-East Asia.

The new product isn’t open-source, and it runs on a range of non-open-source UNIX platforms. The Oracle database system it uses is the antithesis of open-source.

What about this story has anything to do with open source?

The LSE doesn’t think it has anything to do with open source, or even necessarily Linux.

Where this story gets legs among the zealots is that the LSE plans to deploy the new product on Linux, given that the underlying operating system in many cases has been commoditized. Who wouldn’t?

Zealots cheering on trolling columnists like Steven J. Vaughan-Nichols do the profession harm. Now this nonsense is going to be parroted by people who don’t know better, making them look worse for it, for years to come.

I love Linux. I love open source. And you know I also am even quite fond of .NET and SQL Server. I detest fanatics, fanboys, and hysterical columnist that distort or invent reality to get themselves hits.

We have a number of network devices at play in my household.

I have my work machines. My wife has her laptop. My kids have their PCs.

Family and friends that comes to visit connect their iPhones and laptops to our 802.11g/n.

We have shared media that we play on devices like the Xbox 360 and PCs over protocols like uPnP, DAAP and SMB. We have shared files, such as financial info, that we need to access from any machine, all while confident that they remain as secure as we want them to be.

We do VPNing from remote locations like coffee shops to access files and media. We have a shared printer.

To optimize this situation without leaving my power guzzling pig of a PC on around the clock (I’ve setup and torn down a number of home servers over the years, all the ways back to a FreeBSD-based device in the mid-90s that also doubled as a firewall), a while back I bought a Synology DS-106j NAS unit and dropped it on the network, deciding on it after looking at some Windows Home Server products, finding them overpriced and uncompelling (not least because of concern about licensing and DRM BS.)

One of the selling points of this ~$200 unit (to which you add whatever 3.5” hard drive fits your fancy) was the gigabit network port. To me that implied that it must be capable of at least keeping a 100Mbps pipe saturated with goodness.

I also liked that it gently sipped power (13W or so with the hard drive), allaying any guilty about leaving it on around the clock.

The Synology unit is fantastic in many ways.

It has a fantastic interface. It is rock solid (zero downtime with it) and is very feature rich, with several free “firmware” updates since I purchased it adding a number of new features and interface improvements. It is easy to setup and use, and to a superficial degree secure. It comes with great desktop software for common tasks like backing up (and versioning) files on your Windows machines. You can offload torrent downloading to it if you’re into that sort of thing.

But the hardware seems to be underpowered. The 200Mhz MPC8241 CPU that powers it can’t deliver on the extensive feature list, in my opinion.

The 1Gbps network port turned out to be brave talk that it couldn’t back up with action. Normal SMB/CIFS (the protocol that Windows utilizes to access network file shares) averaged about 2.5MB/second throughput with the device, which factoring in overhead is around 22Mbps, leaving plenty of headroom before it hit the limits of even a lowly 100Mbps network.

I got it up to 8MB/second or so by turning off all of the features like media sharing, and then disabling all security on the share and enabling guest access – it seems that access control security significantly adds to the computational burden – but that isn’t a viable day-to-day option so I had to revert and it was back to 2.5MB/second. FTP was a little better, but was still way below expectations given that the same hard drive got a rough 40MB/second from a direct-attached USB 2 enclosure.

Given that I am now downloading from the web at large at 1.1MB/second, this seemed silly. As we’ve started to amass monstrous AVCHD video files of a GB or more from the home video camera (the latest GB or so features my youngest son herding some chickens that had escaped from their pen at a local farmhouse), it was getting unsustainable.

Synology is now selling a purportedly faster unit, but given that I saw much lower real-world speeds than what they graph for the old unit, and their new unit really doesn’t do that much better anyways, it didn’t compel me to upgrade. Many competitive low end home NAS devices have reportedly similar performance in the wild.

And it isn’t just waiting while moving large files or having SyncBack jobs run. Navigating the music directory from a uPnP client made it obvious that the slowness impacted all activities, with basic operations having intolerable delays. Attaching to its network shares from client PCs inexplicably had multi-second waits before responding.

Then again, I’m the guy who thinks that the biggest crisis in the electronics world is the excessive lag before DVD players will eject a disc, so maybe I’m an outlier, but it seemed like a pretty big weakness of what would otherwise be a great product.

It really bothers me when units doing tens or hundreds or thousands of millions of operations per second take perversely long times to do simple things.

On the bright side, the small dimensions of the enclosure made it easy to find it a home, but it also had the S.M.A.R.T. monitoring of the hard drive complaining about borderline heat incessantly. And the printer sharing never worked properly, but given the endless variety of printers I won’t blame that on the NAS unit.

So I decided that it’s time to replace the device, with the following criteria for its successor.

• Relatively cheap, because I’ll probably just be replacing it soon enough anyways
• CIFS and FTP at a minimum
  
• uPnP would be nice. DLNA wouldn’t be bad. DAAP would be gravy
  
• Some method to backup its own files to a remote location, or at worst to an attached USB drive, with security options like encryption
• Good performance. On a switched 1Gbps network I would hope to see read throughput times of 20MB/second or better, and writes not much worse, with a very low latency on activities: Network shares should enumerate close to instantly
• Small enough that it can be physically secured such that the alarm responders would have arrived before a thief could have gotten to and removed it
• Low enough power usage that I don’t have squirrels giving me a beatdown
• A bit of fun. I look for these things to be educational and challenging to a degree. The easiest solution doesn’t provide me the entertainment value, so turnkey is a negative. At the same time, I have a very finite amount of time to deal with things like this, so it can’t require too much coddling

Local RAID isn’t important to me, as availability isn’t a primary consideration (and multiple drives go contrary to power conservation / heat reduction goals, and generally increase maintenance.) Every file that is on the unit always exists on at least one other drive on a separate system, so if the drive died I could quickly rebuild and repopulate and life would be grand again. And it is vastly more likely that two drives in one unit would catastrophically fail than it is that two drives in physically separated devices would fail, especially when the separation between can be thousands of kilometres.

Nor do I want this to function as a media station, which I consider an entirely different function, with an entirely different requirements list. I don’t need or want HDMI output, blu-ray decoding computational power, 3D offloading, a MAME box, or anything of that sort. That’s a different project, with a different set of goals.

After looking at the available market options of targeted NAS devices, and hardware solutions like Shuttle computers, I decided to have a little fun and build my own (in the minimalist stick-some-lego-like pieces together way that is "building" a PC), so I grabbed a couple of components.

• $102.30 - An Intel D945GCLF2 mini-ITX motherboard, featuring an on-board passive cooled Atom 330 dual-core processor. The Atom is not a powerhouse processor by any measure of the imagination, but for this purpose it should be fine.
  
  My only real hesitation with this board is that it uses a relatively power hungry chipset, leading to the mixed-up situation where the chipset has an active fan, yet the CPU has just a heatsink. I’d prefer a lower power board with zero moving parts, but didn’t want to risk going too low on processing power  
• $24.75 - A 2GB DDR2-667 RAM module. Even 2GB is excessive for the targeted purpose
• $55.22 - An Apex MI-008 MINI-ITX Case. The case is small enough that it’s stashable, but big enough that it won’t be a hard drive cooker. Like the motherboard, the case isn't a paragon of efficiency, instead coming with a less efficient power supply that is capable of a grossly excessive 250W, so that isn’t optimal, but I’ll worry about that at a later point
• $126.90 – Western Digital WD10EADS Caviar Green 1TB drive. It is intentionally a bit slower of a drive, putting energy efficiency (and with that reduced heat) ahead of raw speed, but it’s still very fast. This would be a great place to use an SSD drive, so maybe I'll swap to one of those as they mature and become more cost effective for this amount of storage

<$200 for the “NAS enclosure” part, then $127 for a good hard drive with room to grow. Pretty inexpensive, and would probably even make a decent internet appliance, which presumably is the intended use of the motherboard/CPU combo.

Putting it together was a snap, with dutiful oversight from my children, followed by a quick flash to the latest BIOS release.

I installed Ubuntu 8.10 Server from a USB stick (I considered installing FreeNAS, but decided that I wanted something a bit more custom), after which I discovered that the hard drive wasn’t set to bootable and the system wouldn’t boot without the USB key, so I had to boot with the key and use parted to manually set the sda device to the boot device, after which the USB key was no longer necessary.

Installation complete, I stashed it away, sans keyboard or mouse or monitor connection (which it makes no complaints about, happily existing without head, keys or mouse), and access since has been via putty. A magical source and destination of media and files, out of sight and hopefully usually out of mind.

I installed Samba (during the OS install by choosing the file server option), created the users and groups and perms, added some shares, and started accessing it from my workstation. It was all very easy and straightforward, including even setting up AppArmor to enforce mode against the Samba services, adding a small additional guarantee of security.

I started the first file move, ready to witness some low-cost awesomeness.

11MB / second to write to it, averaged out over the copy of a GB sized video file. I got about the same speed reading back from it.

Pretty good compared to 2.5MB/second (and now I had much better security granularity than I had before, on a much more versatile platform), but not quite what I was expecting. Awesomeness not witnessed.

The hard drive isn’t the fastest of the bunch, but it’s still very decent.

I’m far from an expert when it comes to Linux (or more generally Unix) systems, seemingly achieving some small level of localized expertise at intervals in my career, but then the solutions simply work and I don’t have to touch them again, so the knowledge rusts and each time I return I feel like I'm a Linux virgin (who would have ever thought that Linux and virgin would used in the same sentence?) It is similar to my proficiencies with Perl or the equally obtuse Powershell, where I put together very decent solutions, but then several months later look in disbelief that I actually wrote the code that now looks foreign.

Nonetheless, I quickly determined that the network adapter was auto-negotiating itself incorrectly to 100Mbps, and 11MB/second is about the max you could achieve over such a link.

After a quick bit of searching, I added ethtool -s eth0 speed 1000 duplex full autoneg off to my interfaces file (after validating that it worked at the command line). It was now at full 1Gbps glory.

So I performed the file test again.

47MB / second to write to it. 55MB / second to read back from it.

Sweet! That isn’t that far off of local storage, and is entirely with the acceptable zone. It was achieved on a $200 host device. File sharing is hardly the most demanding application nowadays, but I was happy to get something much more usable.

Just to take a side trail here for a moment, I just wanted to mention that Linux, and the evolution and progress of it and related projects, really is amazing. It is astounding that such a solution evolved the way it did. Even people who don’t think they benefit from it regularly use devices run on the OS. The amount of functionality and technology available, including the source so you can roll your own, truly is remarkable. Linux didn’t start this movement, and projects like FreeBSD, sendmail and prior initiatives blazed a successful trail before Linux really took off, but it is the most evident element of that development model.

Back to setting up the homebrew NAS, virtually every step of the way I had to consult the wizard, Mr. Google, but at least the answer was usually only a page 1 or 2 hit away (though Google Groups has taken a perilous dive into close-to-uselessness, and proved of little use.)

To this point I still haven’t had luck finding a good uPnP server for the device. uShare simply didn’t work. Mediatomb inexplicably doesn’t work with the xbox360, and even if it did the fact that its anonymous web user interface has the default behaviour of exposing your entire file system makes me question the wisdom of the developers.

Various other projects didn’t fit the bill or didn’t work for me, and most seemed to have been abandoned.

I thought I’d try the MythTV backend, as it is purported to double as an uPnP AV mediaserver, to find that it bizarrely needs an X server running and an X client on the other end just to perform the setup. I decided to purge it from the system to find it wouldn’t uninstall due to /dev/.static/dev being mounted read only. I found a workaround involving stopping and starting udev, which remounted it as read/write, but at that point the damage was apparently done, and it refused to gracefully leave the system.

So I went MythTV huntin’, committing a moronic error in the process (it tried hard to warn me, but I persisted in my foolishness.) I’ll let the putty log speak to it.

Installing everything once was so much fun that I decided to do it again!

So I reinstalled again. The second time around it was a very quick process (even with a nuked /usr/bin I still had a chance to backup customized config files to a client box before reinstalling, so I really didn’t lose much.)

Power wise, I’ve gone from ~13W for the DS-106j, to 35W to 40W for the new device (measured with a Kill-A-Watt. Handy device.) This represents a pretty big move in the wrong direction, but it wasn’t unexpected. 4-8W or so could likely be saved going to a more efficient, fanless power supply. That’ll be a future experiment. The Caviar Green, at 3-4W, isn’t much more power hungry than an SSD, so that wouldn’t make a huge difference. The motherboard and its chipset is the real piggy of this farmyard.

Alas, in return for the extra power consumption I now have a vastly more interesting network device, serving files at a much more usable pace.

So my first adventure is beginning the work on a .NET-based uPnP server, ensuring that it works on Mono along the way. Thus far it has been nothing but remarkable success, and it still astounds me seeing fairly complex .NET applications running well on a Linux box. The thing really, really works!