The reason for my first home Linux installation was to provide access to the Internet from multiple home computers. It was at time of analog modems and 10Mbit ethernet over coaxial cable. As soon as high speed Internet over ISDN was available (and cheap enough), I changed to that emerging technology. The basic setup of the Internet gateway was a computer with a bunch of IPChains scripts on it. Later ISDN was replaced by ADSL, but in essence the gateway remained the same.

Around seven years ago, I had two servers running at home: a gateway and a fileserver. I never invested in server hardware at home. The hardware was the result of my administration activities – old crappy buggy stuff, bought on ebay, gathered by some customers during replacements etc… On the gateway machine (the smaller one), the self-configured IPTables scipts provided NAT firewalls rules, fetchmail gathered mail for all family members and saved it to a local quickmail installation (I had sendmail running too, but editing m4 files is not my passion), a DNS using Bind 8/9 provided local domain address resolution and DHCP was responsible for configuration of the local computers. I also played around with HylaFax and ISDN4Linux based Phone Answering Machine. It was really funny to have e.G. personalized (caller-id-based) messages for different people. The fileserver machine (the most performant one) was running Samba in PDC mode, and had (at most) two SCSI-U controllers and a doozen old cheap small loud SCSI drives connected to them. In addition, ProFTP server, Apache HTTP Server, MySQL, CUPS (for printing) were also running on the fileserver. I’ve never been a fan of compile-your-linux-yourself approach, but used SuSE distributions. I also played with Fli4L – a German mini Linux to be booted from a floppy. The advantage of that approach was the absence of a hard drive in the gateway, which is very silent then. Later, I also experimented with IPCop, a special Linux platform for the Internet gateway/firewall. The recent two-server-installation consisted of two Intel Pentium III machines, which consumed about 400 Watts of power. The fileserver machine had six SCSI hard drives and four IDE drives in it and weighted about 20 kg.

The first big technological change was the introduction of smart Internet-routers/Gateways by different vendors. So, I got rid of a self-configured NAT-firewall and shifted DNS/DHCP to my fileserver (renaming it just to server). The reason for that is – most DNS/DHCPs built into Internet gateways are silly and can not be compared to Bind name-daemon or ISC DHCP. The advantage of usage a standard device (like e.G. Fritz!Box) is a silent, low-power solution (one device for all instead of 3-5).

From that point on, I started to adjust the number of services running on the server. The first change was a shift from quickmail to postfix and IMAPd (from POP to IMAP). The next change was – from IMAPd at home to IMAP at Google. Now, I’m a massive user of Google Mail – both for gmail.com address and my own domains (see Google for your Domain offer). The next service, that became obsolete on the server was the printing subsystem. It was replaced by a small network printserver (first one from Lexmark, now HP JetDirect). But the real reason for elimintation of it is the independence of CIFS protocol for printing.

The next technological change was the introduction of cheap NAS (Network Access Storage) devices. This allowed to replace 10 old hard drives with one modern located in a small device with 15W power consumption and ethernet socket (like e.G. Buffalo LinkStation). As a result I could move the storage away from the Primary Domain Controller (PDC). The PDC is still needed, otherwise the lookups in “Network Neighborhood” could continue for minutes. In addition, I added some services needed for software developement: especially CVS and SVN repositories and Bugzilla.

As a result I could switch of the last Pentium III off and start up a small Pentium II (a 50W consuming Compaq). It runs DNS, DHCP and Samba for purposes of system infrastructure and Apache HTTPD (with PHP, DAV and Pearl), MySQL, SVN, CVS. In fact, I’m reasoning on buying a SSH access on a Internet-hosted machine (e.G. by domainfactory) and installation of Bugzilla and SVN there. In doing so, I’m following the trend from “IT infrastructure as property” towards “IT as a service”. I was also reasoning about getting rid of the NAS at home, but this is only reasonable with really high speed Internet.

You can find more on that in post just published by me on TechJava…

Don Box

Don Box (here) intoduces an intersting list:

• COM Type libraries
• .NET metadata attributes
• XAML
• OSLO

This means to me, that we speak about a declarative language, with own type system, that allows to develop own types in order to describe entities from my domain. I have no clue, why the link to XAML is present. Maybe, if we speak about a language for data modeling, the representation can be created using XAML.Then Don says, that Microsoft develops a textual internal DSL for modeling. In the same time, he speaks about “visual design” that is applicable to the same model. This seems a little wierd for Microsoft, but is well adopted in Eclipse Ecosystem (e.g with a combination of Xtext and GMF used as two representations linked to the same metamodel). The model repository emerges as a database, which can be compared with Teneo Approach.The next paragraph seems very confusing to me, I’ll try to remove buzzwords out of it: I can write an application by populating the repositry with the definition of my app. Our goal is to make it possible to build apps purely out of data. If not possible, the goal is to make the transition to traditional code easy.

Douglas Purdy

Douglas Purdy is another one who is allowed to natter about OSLO. It seems that the tool chain is pretty ready, because of his wish to use community feedback in order to improve it. Apart of this, he does not reveal any details.

David Chappell

So, I searched a little in the Internet and finally found the following interview with David Chappell. He reveals much more details:

• The language is a designed to create so-called schemas. (or metamodels, how I understood this)
• A schema is something, to that models (schema instances) are conform.
• Both schemas and models (schema instances) are stored in repository.
• There is a way to use the tool (the editor) with both: schemas and schema instances.
• There are some predefined schemas (DSLs, metamodels), which can be used to describe: activities, workslows, systems, etc..
• There are some additional tools that are built upon the DSLs (e.g. process server)

From this point, the OSLO stuff begins to make sense for me. The basic language seems to be a kind of a meta language (meta-meta-model) for definition of DSLs for different purposes. The produced DSLs seems to be defined in a way, that the instances have the same form as the schemas – because of uniform repository and bindings of visual and textual editor. Still, I have no clue how dynamics can be realized.

Darryl K. Tafft

Darryl K. Taft writes on eWeek. Here things gets names, and technical details beging ta make sence. The language is in fact declarative and is called D (has nothing to do with another language called D). Now, after the words of David Chappell, the following text makes more sense: “We’re trying to make it simple to get an idea out of your brain and onto a hard disk” – he is not speaking about a DSL, but about D, which is designed to develop DSLs. Finally, another very important sentence: An Oslo user need not learn the D language to use Oslo, however. “The language is a technical detail for a certain audience,” Lovering said. – this is something clear to me, because you don’t have to understand the DSL definition in order to be able to use a DSL. In addition, if the DSL is both, graphical and textual, you possibly only need to understand the pictures (I argue if the productivity can really be boosted in that way, but this is another story).Then, after I thought that I got the idea, a very strange sentence apears: The Oslo language also is partially based on TLA+, a language developed by Microsoft researcher Leslie Lamport, Lovering said. In fact, I know what TLA+ is, because I used to work with it. TLA+ is untyped formal language for description of dynamics of concurrent systems. It uses actions (arbitary mathematical statements) to describe the change of system behaviour. Leslie Lamport and his team developed a model checker, that can verify temporal and logical properties of the modeled system. The only link I know is tha fact, that Leslie Lamport worked together with David Langworthy, who is working in “Connected Systems Division” and seems to be involved in the development of the D language (he is one of the speaker on PDC).Maybe there is a kind of a action dynamics defined on top of schemas stored in repository. In fact I hope that D is not a behaviour description language.

Mary Jo Foley

Finally, the last source I reviewed for information were several posts ( 1, 2, 3) of Mary Jo Foley.Microsoft is planning to ship a set of predefined schemas within the repository, but customers and software vendors will be able to add their own. – this seems to be the idea of multiple predefined DSLs, because otherwise we will just speak about simple structure of a database.

Summaries

I think in this point of time it does not make sense to gather more sources of OSLO. OSLO seems definitly interesting, and will be published in the end of October. Microsoft entered the modeling arena – this will create a new army of people who understand the words model, DSL and generator. I’m still interesting in the D language: it seams to be declarative, text-based schema definition language with a graphical representation. The defined schemas can somehow be stored in a repository, based on a RDBMS. There are still many open questions like: How to handle dynamics? How to create applications from data? What is the relationship between D and TLA+, I’m familiar with? The best thing is to wait, they will answer the questions soon.