Ten years ago this week, online music pioneer Justin Frankel released a little application dubbed Gnutella that enabled file sharing through a distributed P2P network. Frankel, whose previous claim to fame was programming the then hugely-popular Winamp MP3 player software, supposedly named the client after his favorite hazelnut cream spread, and the first version published online was really more of a proof of concept than anything else.

Still, Gnutella hit a nerve. Napster had been sued three months before, and many file sharers were rightfully fearing that the music industry would eventually prevail in court and force Napster to switch off its servers. With Gnutella, no such switch existed, as the client was allowing direct P2P connections without the help of any centralized server. Add to it the fact that Gnutella, unlike Napster, allowed users to swap videos and software as well as MP3s, and you begin to see why many immediately viewed Gnutella as the next step in P2P file sharing.

A step, one should add, that made Frankel’s employer AOL more than a little nervous. It only took the Internet giant a day to force Frankel and his colleagues to take down Gnutella – but even that was too long, as countless sites quickly started to first mirror, then build upon Frankel’s official Gnutella client. There’s always been a little bit of mystery surrounding the exact happenings of those days, but some people have been musing that a person with a surprising amount of insider knowledge showed up in one of the first IRC chat rooms dedicated to Gnutella soon after AOL pulled the plug, only to provide some very detailed information about the inner workings of the client’s P2P protocol.

Speaking of IRC: Early versions of the software didn’t really have any way for users to connect, save for entering another user’s IP address, which is why IRC quickly became an integral part of the early days of Gnutella. It was also in those IRC chat rooms that the myth of Gnutella as a seemingly invincible P2P protocol was born, and the fact that AOL tried but couldn’t contain the software seemed to fit right into that picture. Gnutella was one of the very first P2P apps I ever wrote about, so I lurked in those chat rooms as well, where people were cheering the fact that someone finally found a file sharing solution that couldn’t be shut down. I still remember one IRC user saying: “We’ve started a damn cult again!”

Only Gnutella wasn’t really ready to be a cult. The network routed search requests from peer to peer, leading to an exponential growth of traffic as its network became bigger. Napster programmer Jordan Ritter described the problem early on in a paper titled “Why Gnutella Can’t Scale. No, Really,” and Frankel himself, who has hardly ever gone on the record about Gnutella, once stated that he was fully aware of “how poorly it would scale” when he released the client.

Still, Gnutella captured the imagination of many, one of them being Mark Gorton, founder of the New York-based Lime Group. Gorton was at the time pursuing a vision of automating businesses through structured data, and Gnutella, as something that could, for example, distribute real estate listings wrapped in XML, seemed to fit that image quite nicely. Early versions of the Gnutella client of Gorton’s LimeWire venture were still written with this vision in mind, hoping to build a P2P network that could eventually be used to do all kinds of things with which we’re now familiar on the web, thanks to web services.

LimeWire’s engineers joined a growing group of developers loosely connected through web sites like the long-defunct Gnutella.wego.com (whose admin Gene Kan tragically committed suicide in 2002) and mailing lists like the one for the Gnutella Developer Forum, and one of the first issues to be tackled was scalability. The introduction of a two-tiered system of ordinary clients and so-called Ultrapeers helped grow both the network as a whole and each user’s search horizon. The idea was also later adopted by the developers of KaZaA, whose own take on this two-tiered approach still lives on in Skype’s P2P network.

Technical improvements like these helped Gnutella to grow, but the competition was quick to catch up. Bram Cohen unveiled a first version of BitTorrent only two years after Frankel had published Gnutella, and BitTorrent quickly became the file sharing client of choice for sharing videos online. Part of BitTorrent’s quick rise to fame was its modular simplicity: Cohen had outsourced much of the search and indexing of files to torrent web sites, only handling the actual distribution of data within the client. Gnutella on the other hand was meant to work without any web server. That made it much more invincible, but also much less accessible to users who migrated from apps and clients to a world of web services.

Another issue that has plagued Gnutella from the beginning is not technical, but legal. The protocol was supposed to outsmart trigger-happy lawyers, but the mere fact that there wasn’t a central switch to turn off the Gnutella network didn’t stop rights holders from going after people and companies associated with it. Lawsuits and legal threats forced Morpheus, Xolox, Bearshare and a number of other companies and developers to throw the towel.

LimeWire got sued by the music industry as well in 2006, but that hasn’t stopped the company from continuing with the development and monetization of its client. LimeWire’s client also utilizes BitTorrent these days, but LimeWire’s VP of Product Management Jason Herskowitz told me during a phone conversation that Gnutella has “worked really well” for the company, and that its engineers are looking into ways to make Gnutella once again more attractive to developers by exposing some of its functionality through web services. “There is still a long future ahead for Gnutella,” he predicted.

Not everyone agrees with that outlook. Adam Fisk, who was hired by LimeWire as one of its first developers in the summer of 2000, but left the company in 2004 to eventually start his own P2P venture dubbed Littleshot, believes that some core assumptions of the Gnutella protocol are outdated. “I don’t think that distributed P2P search makes any sense,” he told me, explaining that the very server-less search functionality that made Gnutella superior to Napster also ended up being its biggest burden, and that it would be much easier to have servers handle search and just use P2P to deliver data – a recipe that has already helped BitTorrent succeed.

Sure, LimeWire and some other Gnutella clients could still stick around for a long time, Fisk admitted, but he was skeptical that we would ever see any significant new project based on Gnutella. “That would be shocking,” he said.

Photo courtesy of (CC BY-SA) Flickr user Jessica  Diamond.

Related content on GigaOm Pro: What’s Next for the Cloud? Distributed Architectures (subscription required)

The last week has seen some interesting progress for the n900. Firstly it was great to see phototranslator finally being availiable in extras-devel. I wrote a couple of weeks ago about having lost patience waiting and played with OCR myself. However phototranslator has put it together in a slick package and combines with google translate api to provide a pretty cool application.
Obviously it is of most use if you are travelling to a foriegn country, translating signs and menus as you go, but it is still interesting to play with and just show off the capabilities of the device (without having to drop to a terminal).

Perhaps the major new item this week is sygic’s mobile maps being released for sale. There was much rumour that they had been waiting to release via Nokia’s ovi store. However, they have made it aviliable for sale via their own site.

I purchased it on Friday after some of the initial rush had died down, and sygic had some chance to get their servers working properly. Some of the first off the mark reported painfully slow downloads which dropped and they had to use resuming downloaders to get all the way through. At 1.8gb I didn’t really want to deal with download problems. Given that the program requires activation via their site, it’s not clear why they didn’t just torrent the file and save their servers a lot of problems.
Nonetheless when I came to make my purchase I got about 200kb/sec and it downloaded in about 90mins.

I had read the maemo forums and seen people had trouble with segfaults if the data folders weren’t in the right place. So I copied to the /home/user/MyDocs folder as instructed. What I didn’t do was unplug my usb cable, just unmounted the n900 and left it charging only. I got a segfault running the application
I rebooted the phone, unplugged the cable and then it ran fine to the point of product activation. Where I selected ‘automatic’ and entered my product code. Only to find it sat doing nothing for a minute then segfaulted….

At this point I was a little concerned about the quality of the app. It’s response to the unexpected seems to be to segfault, which doesn’t seem like good code to me.
However, after actually reading some instructions I realised I should go through manual activation, and that product code != Activation code. I went via their site and got my activation code and at last I was up and running.

Once going, I’ve had no further problems. It’s a fast application and seems very good. I’ve only used it to route me home from work, but it did so well. The thing I noticed was how fast it recalculated when I intentionally deviated from the route. No sooner than I made the turn than I looked down to see new route laid out. To be fair I’m comparing to a now pretty old tomtom, but it’s recalculation always took a few seconds of processing.

The other thing I noticed was that the map has a housing estate in my town that was built perhaps 5 years ago, but does not have some mini roundabouts on my route home from years before that.
I also note that it doesn’t seem to care about traffic lights. By which I mean it gives no indication that it would consider them as a factor in routing decisions. I don’t know if any do, but I hold out hope one day to get routing that knows that 9am on a weekday could mean several extra minutes going through traffic lights.

The maemo forums where quickly full of interesting tips/hacks to enable fullscreen operation & open up more menu options. This allows for portrait operation and more controls. I don’t know why sygic didn’t have these enabled by default, perhaps they are not fully tested so have been left in an implicit ‘unsupported’ state, but easy to switch on.

Some think it’s crazy to pay €59 for something nokia might do for free in ovi maps. However given nokias track record so far I’m not at all convinced they are going to give away anything even close to as good as sygics offering. In terms of price, I paid more for just the France maps addition on my old tomtom, so €59 for the whole of europe seems very good value to me. Now I just need my brodit active holder to ship…

Rumours have increased that firmware release 1.2 is imminent. Based on some wishful thinking and the fact that the UK has finally gotten the 1.1.1 release that the rest of the world got weeks ago. Along with a number of bugs being marked explicitly as in pr 1.2. Neither of these things need have any baring on the release of the next update, but wishful thinking is hard to put down .

I also became aware this week of TweeGo, a new twitter client. This one written with c++/qt and looking very nice. A much slicker ui than my own witter. I am really glad to see more options being actively developed, bringing more choice to n900 users.

Perhaps more significantly than the other things this week… I wrote the code to add avatar support to witter. Though as yet it’s not ready to release, it should be reasonably shortly. (Perhaps this should really be something for next week rather than last)
For a long time I pretty much refused to consider avatar support. I figured it would do nothing but take up memory, use up screen space, and slow things down. And it would cost time in coding I was unwilling to spend. However, this week I came accross a thread on maemo forums with some good examples of what I would need. So I had a play and found it didn’t take too long to get basic support working. Although it must make the memory footprint bigger, it doesn’t appear to hurt performance. So at somepoint soon witter will look something like this:

To have a few new things arriving for the n900 in a short space of time really gave me the feeling that it is gathering momentum. Getting better and better all the time.
This momentum enourages more development, and hopefully more good information such as the thread I found, which in turn leads to more, better applications.

With a cool new technology demo in phototranslator, a great pay-for gps option from sygic and a slick new twitter client in TweeGo, it’s hard not to feel optimistic about the future for the N900 after a week like this.

Hi,

we'd need more customization options for the email notifier. I see on the documentation that TC 5.1 is going to leverage a different templating engine but no additional information is provided.

Our requirements are as following:

• edit the templates from the Web UI, since not every member of the team will have physical access to the TC server.
• reference system properties, environment variables and parameters from the notification template

Is anything of this planned already or should I create an issue on youtrack?

Hi,

we'd need more customization options for the email notifier. I see on the documentation that TC 5.1 is going to leverage a different templating engine but no additional information is provided.

Our requirements are as following:

• edit the templates from the Web UI, since not every member of the team will have physical access to the TC server.
• reference system properties, environment variables and parameters from the notification template

Is anything of this planned already or should I create an issue on youtrack?

Generally, I do not think you should go into such detail and try to workaround issues with your version control using TeamCity. Moreover, the concrete behavior depends on the version control you use.

In general, in case of server-side checkout, TeamCity caches the sources on the server per VCS root checkout settings and goes to the version control server for the clean patch only once per day with all the rest builds only retrieving incremental patch from the VCS server.

So, as far as I understand, if I have several build configurations sharing the same VCS root but each one with different checkout rules (which we set up with the aim of improving checkout performance), do they share a common clean patch or does each one of them require a standalone patch? In the second case I guess our decision to fine-tune the checkout rules per build configuration is counter productive then, and we better remove them completely, can you confirm? In this case, however, we need to configure build triggering rules, since we don't want to see a build starting because something has been checked in which doesn't have anything to do with this build configuration.

However, sometimes building incremental patch can be not quicker then building a full one. e.g. AFAIK for TFS there is no full patch cached by TeamCity.

Do you really need to use non-default checkout directory? This note relates to the case:

I do. Please take into account that certain codebases might be very complex, and while removing the need for a special checkout directory might be very simple in most cases because of well factored code which doesn't need to know where it's running, it other cases, when you have to maintain large chunks of legacy code, this might just not be possible. This is our case, where the code - test code - needs to be running in a specific location on the file system and considering to make it independent of this constraint requires a lot of work.

- a build A with certain VCS settigns is build in the directory X

- whan a build B with different VCS settigns is going to build in the same directory X, TeamCity first deleted the content of the directory and performs clean patch.

BTW, luckily we are in a situation in which different build configurations use the same checkout directory but with same VCS settings.

Read more of this story at Slashdot.

Read more of this story at Slashdot.

Read more of this story at Slashdot.

MF Digital will be Partnering with Disc Makers at the NAB show, April 12-15 to demonstrate CD, DVD, & Blu-ray Disc Duplication, Print and Content Conversion Solutions. (PRWeb Mar 21, 2010)

Read the full story at http://www.prweb.com/releases/2010/03/prweb3754604.htm

FmPro Migrator quickly and accurately migrates FileMaker Pro database structure and data to MySQL, Oracle, Access, SQL Server, Sybase, DB2, OpenBase, PostgreSQL, FrontBase, SQLite and Valentina.

• Converts Access Forms/Reports, Relationships, Value Lists, Queries and Visual Basic to FileMaker 10
• Converts FileMaker Pro Layouts/Scripts/Value Lists to Microsoft Access
• Converts FileMaker Pro Layouts/Scripts/Value Lists, Relationships to Servoy Projects
• Converts Layouts to HTML/JavaScript
• Converts Layouts/Scripts to Revolution Stacks
• Migrates Access, Bento, Firebird and SQL Server to FileMaker 10
• Converts Bento Forms to FileMaker Layouts
• Automated Table Consolidation Feature
• Generates Perl CGI scripts for all databases
• Facilitates FileMaker ESS Configurations
• Host FileMaker Pro data at any ISP supporting MySQL
• Documents FileMaker Pro database structure
• Migration of large text fields and image data for all supported databases
• No ODBC driver licensing required for UNIX servers
• Repeating Fields Extraction
• Economical Single-User License
• FileMaker image export feature
• Export Layout XML to individual files (Developer Edition Feature)

FileMaker Pro database fields containing more than 255 or 4000 characters of text are migrated to the appropriate large text column type in the destination database. Container field data within FileMaker Pro container fields is migrated directly to the appropriate BLOB type of database columns through a network connection. No manual data entry is required for processing either of these specialized data types.
WHAT'S NEWVersion 5.57:

• Added new PHP to revTalk conversion feature.
• Added new BASIC to revTalk conversion feature.
• Fixed the Value Lists generated for Servoy migration projects to add rn instead of only r as value list item separators.
• Updated Fmig_Preferences.db3 (v6) and MigrationProcess.db3 (v4) files to include new preference items for new features.
• Renamed Convert Database to Revolution feature with new icon and changed name to Convert Layouts to Rev.
• Removed obsolete Perl CGI code generating feature.

• Mac OS X 10.5 or later.
• FileMaker.
• ODBC Driver.

• Top Queries
• Top Users
• Top Tables
• Replication Profiling
• Zoomable GUI
• Free Version

• Added reminder message when license is about to expire.
• Fixed bug #67: Software updates fails on Windows in some situations, related to slow network responses.
• Fixed bug #68: Top Tables tab wont always show all tables that are being used. This can happen if the tables have very low usage.
• Fixed bug #72: Memory leak in pie chart painting leading to OutOfMemoryErrors during prolonged running (days).

• Mac OS X 10.5.2 or later
• Intel 64 bit processor
• Java 6)

Read more of this story at Slashdot.

Read more of this story at Slashdot.

John Gilmore keynote – What do we do next, having produced a free software system for our computers? Perhaps we should aim at Windows? Wine + an extended ndiswrapper to run other hardware drivers + a better system administration interface/resources/manuals. However that means knowing a lot about windows internals – something that open source developers don’t seem to want to do. We shouldn’t just carry on tweaking – its not inspiring; whats our stretch goal? Discussion followed – reactos, continue integrating software and people with a goal of achieving really close integration: software as human rights issue! ‘Desktop paradigm needs to be replaced’ : need to move away from a document based desktop to a device based desktop. Concern about the goal of running binary drivers for hardware: encourages manufacturers to sell hardware w/out specs; we shouldn’t encourage the idea that that is ok. Lots of concern about cloning, lots of concern about what will bring more freedom to users, and what it will take to have a compelling vision to inspire 50000 free software hackers. Free software in cars – lots of safety issues in .e.g brake controllers, accelerators.

Eben Moglen – ‘We’re at the inflection point of free software’ – because any large scale global projects these days are not feasible without free software. Claims that doing something that scales from tiny to huge environment requires ‘us’ — A claim I would (sadly) dispute. Lots of incoming and remaining challenges. ‘Entirely clear that the patent systems relationship to technology is pathological and dangerous’ – that I agree with! Patent muggings are a problem – patent holders are unhappy with patents granted to other people . Patent pools are helping slowly as they grow. Companies which don’t care about the freedom aspect of GPLv3 are adopting it because of the patent protection aspects. Patent system is at the head of the list of causes-of-bad-things affecting free software. SFLC is building coalitions outside the core community to protect the interests of the free software community. We are starting to be taken for granted at the high end of mgmt in companies that build on free software. … We face a problem in the erosion of privacy. We need to build a stack, running on commodity hardware that runs federated services rather than folk needing centralised services.

Marina Zhurakhinskaya on GNOME Shell: Integrates old and new ideas in an overall comprehensive design. Marina ran through the various goals of the shell – growing with users, being delightful, starting simply so new users are not overwhelmed. The activities screen looks pretty nice The workspace rearrangement UI is really good. The notifications thing is interesting; you can respond to a chat message in-line in the notification.

Richard Stallman on Software as a Service – he presented verbally the case made in the paper. Some key quotes… “All your data on a server is equivalent to total spyware” – I think this is a worst-case analogy; it suggests that you can never trust another party: kindof a sad state of paranoia to assume that all network servers are always out to get you all the time. And I have to ask – should we get rid of Savannah then (because all the data is stored there) – the argument for why Savannah is not SaaS is not convincing: its just file storage, so what makes it different to e.g. Ubuntu One? “If there is a server and only a little bit of it is SaaS, perhaps just say don’t worry about it – because that little bit is often the hardest bit to replace.”  ”Lets write systems for collaborative word process that don’t involve a central server” — abiword w/the sharing plugin ? RMS seems to be claiming that someone else sysadmining a server for you is better than someone else sysadmining a time-shared server for you: I don’t actually see the difference, unless you’re also asserting that you’ll always have root over your’ own machine’. The argument seems very fuzzy and unclear to me as to why there is really a greater risk – in particular when there is a commercial relationship with the operator (as opposed to, say, an advertising supported relationship).

Canonical has announced the availability of the first Ubuntu 10.04 beta release. The new version of Ubuntu, codenamed Lucid Lynx, is scheduled to arrive in April. It will be a long-term support (LTS) release, which means that updates will be available for three years on the desktop and five years on servers.

Although the Ubuntu developers have largely focused on boosting stability for this release, they have also added a number of noteworthy new features and applications. One of the most visible changes is the introduction of a new theme—a change that is part of a broader rebranding initiative that aims to update Ubuntu's visual identity.

Read the comments on this post

TeamViewer connects to any PC or server around the world within a few seconds. You can remote control your partner's PC as if you were sitting right in front of it.

eToolz includes some of the most important networking tools like NS-Lookup, Ping, TraceRoute and Whois. It allows also to check E-mail addresses (MX-Server) and to search/convert domain names showing DNS entries (A, PTR, NS, SOA). eToolz supports also international domain names.

Note: the application starts in German, but you can immediately focus the "Sprache / Language" option in the window.

BlackBerry addicts that keep strange hours should keep a stiff drink on hand between 2AM and 6AM EDT on Sunday, March 28th, because those are the hours RIM will be hard at work updating its North American servers to BIS 3.0. During this four hour upgrade process literally every BIS user will have to go without BlackBerry services, but since it’s happening so late in the East, it’s going to be the poor ol’ West Coasters that are going to feel the brunt of this. Either way you slice it, the gain is worth the pain as BIS 3.0 brings with it 2-way Gmail sync (in the EMEA region this is being slowly introduced, so take that as a sign the same will happen here), the ability to add/delete Gmail labels, support for .wma and OpenOffice files, and a few other assorted trinkets that concern carrier-related error messages. Sweet.

Read

Ten years ago this week, online music pioneer Justin Frankel released a little application dubbed Gnutella that enabled file sharing through a distributed P2P network. Frankel, whose previous claim to fame was programming the then hugely-popular Winamp MP3 player software, supposedly named the client after his favorite hazelnut cream spread, and the first version published online was really more of a proof of concept than anything else.

Still, Gnutella hit a nerve. Napster had been sued three months before, and many file sharers were rightfully fearing that the music industry would eventually prevail in court and force Napster to switch off its servers. With Gnutella, no such switch existed, as the client was allowing direct P2P connections without the help of any centralized server. Add to it the fact that Gnutella, unlike Napster, allowed users to swap videos and software as well as MP3s, and you begin to see why many immediately viewed Gnutella as the next step in P2P file sharing.

A step, one should add, that made Frankel’s employer AOL more than a little nervous. It only took the Internet giant a day to force Frankel and his colleagues to take down Gnutella – but even that was too long, as countless sites quickly started to first mirror, then build upon Frankel’s official Gnutella client. There’s always been a little bit of mystery surrounding the exact happenings of those days, but some people have been musing that a person with a surprising amount of insider knowledge showed up in one of the first IRC chat rooms dedicated to Gnutella soon after AOL pulled the plug, only to provide some very detailed information about the inner workings of the client’s P2P protocol.

Speaking of IRC: Early versions of the software didn’t really have any way for users to connect, save for entering another user’s IP address, which is why IRC quickly became an integral part of the early days of Gnutella. It was also in those IRC chat rooms that the myth of Gnutella as a seemingly invincible P2P protocol was born, and the fact that AOL tried but couldn’t contain the software seemed to fit right into that picture. Gnutella was one of the very first P2P apps I ever wrote about, so I lurked in those chat rooms as well, where people were cheering the fact that someone finally found a file sharing solution that couldn’t be shut down. I still remember one IRC user saying: “We’ve started a damn cult again!”

Only Gnutella wasn’t really ready to be a cult. The network routed search requests from peer to peer, leading to an exponential growth of traffic as its network became bigger. Napster programmer Jordan Ritter described the problem early on in a paper titled “Why Gnutella Can’t Scale. No, Really,” and Frankel himself, who has hardly ever gone on the record about Gnutella, once stated that he was fully aware of “how poorly it would scale” when he released the client.

Still, Gnutella captured the imagination of many, one of them being Mark Gorton, founder of the New York-based Lime Group. Gorton was at the time pursuing a vision of automating businesses through structured data, and Gnutella, as something that could, for example, distribute real estate listings wrapped in XML, seemed to fit that image quite nicely. Early versions of the Gnutella client of Gorton’s LimeWire venture were still written with this vision in mind, hoping to build a P2P network that could eventually be used to do all kinds of things with which we’re now familiar on the web, thanks to web services.

LimeWire’s engineers joined a growing group of developers loosely connected through web sites like the long-defunct Gnutella.wego.com (whose admin Gene Kan tragically committed suicide in 2002) and mailing lists like the one for the Gnutella Developer Forum, and one of the first issues to be tackled was scalability. The introduction of a two-tiered system of ordinary clients and so-called Ultrapeers helped grow both the network as a whole and each user’s search horizon. The idea was also later adopted by the developers of KaZaA, whose own take on this two-tiered approach still lives on in Skype’s P2P network.

Technical improvements like these helped Gnutella to grow, but the competition was quick to catch up. Bram Cohen unveiled a first version of BitTorrent only two years after Frankel had published Gnutella, and BitTorrent quickly became the file sharing client of choice for sharing videos online. Part of BitTorrent’s quick rise to fame was its modular simplicity: Cohen had outsourced much of the search and indexing of files to torrent web sites, only handling the actual distribution of data within the client. Gnutella on the other hand was meant to work without any web server. That made it much more invincible, but also much less accessible to users who migrated from apps and clients to a world of web services.

Another issue that has plagued Gnutella from the beginning is not technical, but legal. The protocol was supposed to outsmart trigger-happy lawyers, but the mere fact that there wasn’t a central switch to turn off the Gnutella network didn’t stop rights holders from going after people and companies associated with it. Lawsuits and legal threats forced Morpheus, Xolox, Bearshare and a number of other companies and developers to throw the towel.

LimeWire got sued by the music industry as well in 2006, but that hasn’t stopped the company from continuing with the development and monetization of its client. LimeWire’s client also utilizes BitTorrent these days, but LimeWire’s VP of Product Management Jason Herskowitz told me during a phone conversation that Gnutella has “worked really well” for the company, and that its engineers are looking into ways to make Gnutella once again more attractive to developers by exposing some of its functionality through web services. “There is still a long future ahead for Gnutella,” he predicted.

Not everyone agrees with that outlook. Adam Fisk, who was hired by LimeWire as one of its first developers in the summer of 2000, but left the company in 2004 to eventually start his own P2P venture dubbed Littleshot, believes that some core assumptions of the Gnutella protocol are outdated. “I don’t think that distributed P2P search makes any sense,” he told me, explaining that the very server-less search functionality that made Gnutella superior to Napster also ended up being its biggest burden, and that it would be much easier to have servers handle search and just use P2P to deliver data – a recipe that has already helped BitTorrent succeed.

Sure, LimeWire and some other Gnutella clients could still stick around for a long time, Fisk admitted, but he was skeptical that we would ever see any significant new project based on Gnutella. “That would be shocking,” he said.

Photo courtesy of (CC BY-SA) Flickr user Jessica  Diamond.

Related content on GigaOm Pro: What’s Next for the Cloud? Distributed Architectures (subscription required)

L'hébergement virtuel proposant des "tranches" de serveur devient de plus en plus populaire. Vous trouverez un peu partout des offres comparables à celles proposées par Gandi ou Slicehost et, de mon point de vue, elles représentent l'avenir de l'hébergement de sites web.

En regardant les offres de près, le premier réflexe est généralement d'être effrayé par les tarifs. Surtout en France, où certaines sociétés laissent croire qu'on peut fournir un hébergement de qualité pour 2 EUR par mois.

Pour une somme allant de 14 à 20 EUR par mois, vous pourrez trouver un hébergement disposant de 256 Mo de RAM, un puissance CPU raisonnable et une flexibilité que vous ne trouverez pas ailleurs. Et vous serez administrateur de votre machine ce qui n'est pas le dernier des avantages.

Il y a de cela 8 ans, j'avais un vieux PC dans ma cuisine, avec 192 Mo de RAM, un CPU poussif et un débit montant de 128 kbits/s. Ça ne m'empêchait pas de faire tourner un serveur web, un serveur de mail. Les début de dotclear.net se sont faits sur cette machine, à côté du frigo.

Du coup, je me suis dit, qu'il n'y avait aucune raison de ne pas faire fonctionner correctement un serveur web avec 256 Mo de RAM, un meilleur CPU et une bande passante bien plus confortable.

Afin de tester mon idée, j'ai créé une machine virtuelle avec vmware et une image Ubuntu server. Cette liste propose un grand nombre d'images de VM prêtes à l'emploi, ça vous fera gagner du temps si vous voulez faire des essais.

Au démarrage la machine virtuelle n'utilise que 16% de ses 256 Mo de mémoire. (Avec uniquement un serveur SSH).

Sur Ubuntu, la configuration par défaut de MySQL est réputée pour consommer un peu trop de ressources. Nous pouvons améliorer ceci en modifiant un peu le fichier /etc/mysql/my.cnf. Faites en sorte d'avoir ces lignes dans le fichier :

Si vous n'avez pas l'intention d'installer Dotclear ou plus généralement, ne comptez pas utiliser InnoDB, désactivez le en ajoutant la ligne :

Cette configuration me donne une utilisation de mémoire d'environ 30 Mo. Ça dépendra, bien sûr, de vos applications.

De l'ensemble LAMP, nous voulons conserver Linux, MySQL et PHP. Nous allons remplacer Apache par un autre serveur : Nginx. Nginx (on dit Engine X) est ce qu'on pourrait appeler la nouvelle génération de serveurs web. Il peut remplacer Apache dans de nombreuses circonstances. Je vous laisse lire la présentation du serveur ainsi que ce billet.

Pour installer Nginx, apt, yum ou votre gestionnaire de paquet feront l'affaire. Assurez vous simplement d'avoir une version 0.7.x. Nous verrons ensuite comment le configurer.

Nginx, contrairement à Apache, ne propose pas d'équivalent à mod_php. Vous ne pouvez pas non plus utiliser PHP simplement en CGI (tant mieux). Vous allez devoir utiliser FastCGI et indiquer à Nginx comment y accéder.

Pour faire simple, nous allons installer spawn-fcgi et écrire un script pour lancer nos processus PHP. Sur Ubuntu/Debian :

Le script permettant de lancer les processus PHP est le suivant, vous devez le créer dans /etc/init.d/php5-fcgi :

Pour activer votre script (avec Ubuntu ou Debian) :

Maintenant vos processus PHP se lanceront au démarrage. Un petit mot sur deux paramètres importants. Les processus PHP en CGI ont une tendance connue à planter de manière régulière. C'est à ceci que sert à la variable PHP_FCGI_MAX_REQUESTS que nous avons mis à 500. Tous les 500 cycles, chaque processus PHP sera relancé. Enfin, PHP_FCGI_CHILDREN=2 indique de lancer deux processus PHP. Vous pouvez en mettre plus mais n'oubliez pas que plus il y en a plus vous consommerez de mémoire.

Une fois en production, vous pourriez constater que c'est un peu lent. L'installation du paquet php5-xcache améliore grandement les performances de PHP. Consultez le site de Xcache pour en savoir plus. J'ai doublé la taille du cache (xcache.size dans /etc/php5/conf.d/xcache.ini) mais ne perdez pas de vue que cette taille est appliquée à chaque processus PHP (soit 2 fois 32M dans mon cas).

Maintenant, vous pouvez lancer votre processus à l'aide de /etc/init.d/php5-fcgi start.

Notez également que PHP FPM peut remplacer spawn-fcgi. J'en ai lu le plus grand bien mais ne l'ai pas testé.

Vous avez bien entendu installé Nginx avec un classique apt-get install nginx.

Nginx pour Ubuntu ou Debian fourni un fichier de configuration pour FastCGI. Nous allons le compléter. Le fichier est /etc/nginx/fastcgi_params et doit contenir ceci :

Les derniers paramètres permettent d'utiliser PHP en mode PATH_INFO (à condition de bien configurer Nginx). Le tout dernier paramètre est le plus important, c'est celui qui va dire à PHP quel script exécuter.

Maintenant, testons ceci. Créez un fichier /var/www/nginx-default/test.php dans lequel vous pouvez par exemple mettre un appel à phpinfo() ou ce que vous voulez.

Ensuite, modifiez le fichier /etc/nginx/site-enabled/default (il peut se trouver ailleurs si vous n'êtes pas sur Ubuntu/Debian). Vous pouvez aussi créer un nouveau fichier, vous faites comme vous voulez.

Relancez le serveur avec /etc/init.d/nginx restart et rendez vous sur votre nouveau site pour pointer sur le fichier test.php. Votre script doit s'exécuter.

Ça ne marche pas ? Vous obtenez le très informatif "No input file specified" ? Quelque chose est mal configuré. Ça peut être un problème de permission. Commencez par essayer d'enlever la partie se chargeant d'interpréter PHP et de charger votre script. Vous pouvez également utiliser strace pour repérer quelle valeur de SCRIPT_FILENAME est passée à PHP.

Dotclear, avec la configuration que nous venons de voir, s'installe très bien. Vous pourrez même l'utiliser en PATH_INFO avec une URL du genre /index.php/...

Peut-on faire mieux ? Yes we can.

Avec Apache, vous aviez la possibilité d'utiliser Mod Rewrite pour lui dire quelque chose comme : "Si ce n'est pas un fichier ou un répertoire transforme l'URL vers tel script". Nginx propose des options de rewrite du même type avec une limite de taille ; on ne peut pas imbriquer les tests. On peut donc tester si l'URL demandée pointe vers un fichier, ou vers un répertoire mais pas les deux. C'est gênant. On peut s'en sortir en écrivant une configuration très complexe et difficile à maintenir. Sinon on peut se rappeler que Nginx n'est pas Apache et qu'il existe peut-être un autre moyen.

Nginx fournit une directive appelée try_files qui va faire exactement ce que nous ferions avec Apache rewrite en testant si le fichier et le répertoire n'existe pas.

Donc, pour votre Dotclear à la racine pointant vers /dotclear/index.php, voici la recette :

La même chose est possible pour Wordpress, Drupal et consort.

Ma configuration est une machine virtuelle VMWare avec 256 Mo de RAM utilisant un seul core de CPU sur un MacBook Pro. C'est sans doute un peu plus que ce dont dispose une part d'hébergement mais ça donne une idée.

Voici quelques chiffres :

C'est tout à fait honorable pour du PHP avec aussi peu de ressources. L'autre aspect intéressant est que même avec 50 visiteurs au même moment, vos fichiers statiques sont servis en moins d'une demie-seconde. Et contrairement à Apache, l'utilisation mémoire n'augmente pas avec le nombre de requêtes simultanées.

La conclusion est que, oui, avec 256 Mo de RAM, vous pouvez disposer d'un serveur web tout à fait performant et vous disposez même d'une marge de manoeuvre permettant d'installer un serveur SMTP et IMAP.

N'étant pas non plus un expert sur le sujet, n'hésitez pas à signaler les erreurs. Et si vous avez des astuces concernant PostgreSQL avec une telle configuration, ça m'intéresse.

At the MIX10 conference this past week, MAKE had a booth embedded inside the Coding4Fun exhibit. We showed off MAKE magazine, some of our Make: Books, and a few kits. Our Coding4Fun neighbors had a lot of great projects on display, including their Coding4Fun Cannon, which was a robotic t-shirt launching platform with two barrels. What's more, it was controlled by a Windows Phone 7 app, and they completed it in two weeks. Here's how Clint Rutkas of Coding4Fun described it:

Read all about it at Coding4Fun: Coding4Fun Cannon - Project Overview
Read a sample chapter from the upcoming Learning Windows Phone Programming at O'Reilly Answers