"www. is deprecated"?

[pne]

scribot_feed linked to no-www.org just now, which reminded me—I think I had come across that site before but had forgotten what it was called.

Personally, I disagree with their analysis of the situation; it appears to hinge on a disagreement we have over terminology.

Specifically, they talk about "www" as a "sub-domain" with language such as Why then do many servers require their websites to communicate through the www subdomain?. I tend to call the first "label" of a domain name the "host name" (for want of a better word, since the full name is also a "host name" in my book, as well as a "domain name"). So in "www.livejournal.com", "www" would be the "host name" and "livejournal.com" would be the domain. It's a host inside a domain.

Using http://livejournal.com/ seems weird to me since it appears to be addressing not a single host but a whole domain—even though my browser connects to a certain host, a single machine (possibly one of many sharing the same name in load-balancing situations, but one machine for any given connection). Connecting to a "domain" seems… wrong to me.

They also mention that Mail servers do not require you to send emails to recipient@mail.domain.com. but that's a red herring in my eyes: mail servers do not require this specifically because of how email works with DNS.

In the olden days (as far as I know), mail to user@example.com would cause the mail transfer agent to attempt to connect to the machine example.com, but we've had MX (mail exchanger) records for ages now, and they're the preferred method for determining which host to send email to.

So when you want to send mail to user@example.com, the mail transfer agent now looks up example.com's mail exchangers in the DNS. Those might be, say, mail.example.com, backup-mx.example.com, and othermx.otherhost.com (i.e., they could even be in a completely different domain name!); then, according to certain priority rules, it will connect to one of those hosts to attempt to deliver the message. Only if no MX records can be found for a domain will the MTA, for historical/compatibility reasons, attempt to connect to a host whose name is the portion after the @ sign.

So the reason why we don't need to write "recipient@mail.domain.com" (if mail.domain.com is the mail exchanger for domain.com) is because of MX records in DNS. But there's no concept of "HTTP exchanger" in DNS; I vaguelly recall a concept of service records (RR SVC?) which could do this (i.e. say that HTTP traffic to domain X is handled by a host named Y), but I don't think that's widely supported.

Historically, a URL of http://FOO for any FOO means to browsers "convert 'FOO' into an IP address by looking up an 'A' record for that name, then connect to that IP address (or one of them, if there are multiple 'A' records". Whereas mail sent to user@FOO means to MTAs "lookup hostnames by finding 'MX' records for 'FOO' and order them by priority. Then take the 'best' MX (with the lowest priority number) and convert that hostname to an IP address by looking up an 'A' record for that name, then connect to that IP address". To rather different approaches.

Where subdomains come in is in names such as www.hpl.hp.com, which for me is host "www" in the "hpl" (HP labs) subdomain of the "hp.com" domain. But in www.example.com, I wouldn't call "www" a subdomain.

If people say that calling the host containing their websites "www" is redundant because you're accessing it with a URL scheme of "http", I can accept that—use a different hostname, such as "web" if you'd like (as MIT did early on, with web.mit.edu being the host with the school's official website, and www.mit.edu coming later and hosting a student organisation) or something completely different such as "argon" or "senator-bedfellow" or "salticus-peckhamae". Call your main website http://marvin.example.com/ if you don't like "www". But give the machine a name of its own, by which it will be known in the domain, and construct the URL by tacking "." + the domain onto the machine name! http://example.com just seems weird to me, like a person without a given name.

---

I wonder whether the idea that "example.com" can be the name of a machine comes from people accustomed to "Internet" (in the sense of TCP/IP internetwork) = "World Wide Web", and the many domains where they only see IP traffic going to and from a web server, which is often not even on a machine of its own but on a shared webhoster. In which case, I suppose it makes a bit of sense to consider vanityhost.com as the label of the machine since there is only one machine in that domain and it exists to serve web traffic.

On the other hand, I grew up with the model of a domain being a bunch of machines, say, in an office: several dozen workstations, each with a name, a file server or three, a database server, oh, and also a web server. All the machines being something.example.com, and often being able to call each other simply something since .example.com is the default domain for traffic inside. But each host has a name (not merely an IP address) of its own: one single word, which can be disambiguated on the Internet by adding on the name of the domain.

Comments

[crschmidt.livejournal.com]

In mail, MX records point to the main server.
In Web, the main DNS record points to the main server.

There can be reasons for http://crschmidt.net/ and http://www.crschmidt.net/ to have different content: if one is offering a list of services, and one is specifically web related, for example. However, given the fact that most domains don't have a specific "web" or "www" portion, there is no need to specifically address the www part - and assuming that this is the case, there's no reason that the default resolution of example.com shouldn't point to the website.

www is redundant, because you're requestion port 80 over HTTP. If you're accessing a different service - svn.crschmidt.net, or trac.crschmidt.net, or mail.crschmidt.net, or athena.crschmidt.net - then it makes sense to identify it. But just like the "Default" for mail is to look up the MX record, the default for the web is to look up the IP address associated with the domain name, and if that IP isn't going to serve a different purpose, there's no reason not to make that the primary location of the WWW info.

I'm a no-www supporter.

[pne.livejournal.com]

the default for the web is to look up the IP address associated with the domain name, and if that IP isn't going to serve a different purpose, there's no reason not to make that the primary location of the WWW info.

See, I feel weird assigning an IP address/an A record to a domain name in the first place. For me, domain names are containers which contain hosts, and hosts have IP addresses. Domain names don't have IP addresses. The records I'd expect to find for a domain are NS, SOA, and MX, but not A or AAAA or CNAME.

Sort of like how you have Philip Newton and Stella Newton but nobody called "Newton". An A record for crschmidt.net seems to me like a name "Newton" identifying a person -- for me, "Newton" identifies a family and "crschmidt.net" identifies a domain. Families have people, each of whom has a personal name; domains contains hosts, each of which has a personal name (and an IP address). Telling a browser to request http://crschmidt.net/ seems to me like having someone knock on my door and ask for Newton.

Or for a marginally better analogy, having someone phone and ask for Newton. I do most of the phone traffic in our family (at least outbound), so I might be the "default phone service". But that doesn't mean that "for phone purposes, 'Newton' refers to me"; I still prefer to be "Philip Newton".

I guess I have a different conceptual model of what a domain is that makes me see example.com referring to a specific machine as "wrong". I like to think that my way of thinking is the traditional way, though.

www is redundant, because you're requestion port 80 over HTTP.

Go ahead and call the machine something else -- I still think the machine should have a name of its own, though.

Having "www" be the default (through convention/tradition) name of the machine means that it's easy to guess what the FQDN of the main web presence for a company is, given its domain name, something that would be more difficult if people used whatever hostnames they liked for their main webserver (and didn't alias them with CNAME).


... Do you understand where I'm coming from, though, even if it's not the model you have?

[nik-w.livejournal.com]

I agree with your thoughts on all of this - I do sometimes think www is a little redundant, but I still prefer to see it there.

> Or for a marginally better analogy, having someone phone and ask for Newton.

That happened here today - the bank phoned up and asked for me by my surname. However, being as it was my dad that answered, he assumed it was for him. Confusion arose when he managed to get all the security questions wrong (inlcuding "month of birth" - which is pretty hard to get wrong:P). My dad eventually figured it out when they asked for his mobile phone number and told him that was wrong too!:p If they had only asked for his first name (or their computer system was smart enough to notice that there was several people with the same surname that banked with them from our house), then there'd have been much less confusion!

[crschmidt.livejournal.com]

... Do you understand where I'm coming from, though, even if it's not the model you have?

Yep. I just disagree.

The machine does have a name of its own: the fact that you don't know it is irrelevant. The machine is named 65.110.51.60: crschmidt.net is just a shortname for it. Mail servers have their own names too, but again, you have no need to know it. Then again, my mail server is set up to think it's at crschmidt.net too, so I don't fit your model at all. :)

[darth-spacey.livejournal.com]

I agree entirely.