Look back in sorrow. Modern tech's cool, but many things used to be so very much better than now.

[liam_on_linux]

I love my Android phone in some ways - what it can do is wonderful. The formfactor of my Nokia E90 was better in every single way, though. Give the Nokia a modern CPU, replace its silly headphone socket, MiniUSB port & Nokia charging port with a standard jack & a MicroUSB, make the internal screen a touchscreen, and I would take your arm off in my

Comments

[liam_on_linux]

I am no Jon Postel or Vint Cerf. I do not know how the upgrade of IPv4 could have been done better. However, 6 was spun as being simpler than 4 and the reverse is actually true.

Given that there were routable WAN protocols such as DECnet and IPX 30+y ago, I feel that there must be better answers than those that we have.

Yes, the current situation with IPv4 /is/ a mess and is not sustainable. However, I personally find the reluctance to move to IPv6 entirely understandable. I spent weeks trying to wrap my head around it and largely failed. IP only took off when it got to the stage where installation consisted of:

[1] Tick the TCP/IP box
[2] Enter an address
[3] Enter the server address
[4] Click "done".

Later, that got streamlined, with DHCP and stuff. (One of the few things for which MS deserves serious credit, IMHO!) WINS also worked quite well, for all that it is demonised now. I sorted all a major City bank's network problems with a few clicks and a WINS server, a decade ago, and was the hero of the hour; none of their other network wranglers had even thought of it. I suspect they didn't know what it was. Even today, for many network techs, my ability to calculate a simple netmask in my head /and apply it/ (to very simple addresses!) is regarded with awe, as sorcery of a high order.

Most people are not very bright. IT is now a mundane job, done by people in it for the money, with no interest or passion, who therefore know as little as they can get away with. They do not know what binary /is/ let alone what a "mask" is or how to apply one. These are magic incantations.

And yet these are the people designing and running the systems that hold our lives in their mechanical claws: our bank balances, our life-support systems, the fly-by-wire control systems in our vehicles.

It needs to get a /lot/ simpler. Instead, it is getting harder. The big fat manuals are now DVDs or websites, plus a thriving 3rd party market, but they are fatter and hairier than ever.

- - - - -

As for NetBEUI, yes, in the late 1980s, it was actually a wire protocol. Wikipedia does not cover this, last I checked. It was the default network protocol of the MS Network for DOS, 3Com 3+Share, 3Com 3+Open (the OS/2-based successor) and Windows for Workgroups.

Now, the term is used for a level in the stack. Once it was an entire protocol. The single name refers to multiple different things.

C.f. DLC: once an IBM network protocol, later used for networked HP printers. Now means "downloadable content" in games.

[steer]

However, 6 was spun as being simpler than 4 and the reverse is actually true.

You should listen to the guys who invented IPv6 not the guys who wrote the press releases (not that there were any). IPv6 was the smallest possible change in IPv4 that could fix address exhaustion and help out with things like anycast and slightly with mobility. However, at the cost that deployment would be a nightmare. Deploying something which was more of a change would be more of a nightmare.

However, I personally find the reluctance to move to IPv6 entirely understandable.

Don't bother moving to it. I'm not going to until the rest of the world does. Indeed I've got a long term bet with -bat. that IP will finish as the ubiquitous network protocol before IPv6 reaches high rates of deployment. This isn't the fault of IPv6 though... it's the smallest change that could possibly be made.

Instead, it is getting harder.

Really? Because today I'm pretty confident I can code up a wiki from scratch in a day. I can code up something which scales out to arbitrarily many users in a week starting from no hardware and purchasing no hardware. I no longer have to bother with plugging wires in machines or finding out about IP addresses.

Things are getting easier. Much easier. It's just the tools you use have changed vastly. If you're finding it harder nowadays it's because you're using the wrong tools.

[steer]

Now, the term is used for a level in the stack. Once it was an entire protocol

A protocol lives at a level in the stack. TCP is a protocol. It lives at level 4. IP is a protocol. It lives at level 3. MPLS is a protocol it lives at level 2.5. As I understand NetBEUI it was level 3 with no proper level 4. May be wrong.

[liam_on_linux]

The OSI 7-layer model is a useful tool for thinking about how networks work, but few networks have actually been implemented using actual discrete versions of the OSI's layers. ISTR reading of some in the '90s but they bombed hard AFAIK.

Between the era of Microsoft DOS network clients, original for the IBM PC Network and continuing up to the era of Win9x, "NetBEUI" was the default wire protocol used by MS-DOS, OS/2 and Windows 3/95/98/NT.

Wikipedia claims it was misnamed: http://en.wikipedia.org/wiki/NetBIOS_Frames_protocol

This may be so, but referring to it by a more accurate name that no actual product ever used and which no other reference I've ever seen uses is, to my mind, unhelpful to the point of uselessness.

As Wikipedia, seemingly reluctantly, acknowledges:
... it has such an advantage over TCP/IP that requires little configuration.

That is precisely the point that I am getting at.

In the 1980s and early 1990s, building a LAN meant plugging the nodes into a shared wire and making sure they all had a network stack. Later, the notion of /naming/ the network came along.

This was replaced by a system which requires, as a minimum:

* a specifically-assigned unique ID for every node - different from the unique ID already provided by its network hardware
* or the installation of a server to dispense said IDs automatically, from a pool which must be manually provisioned, over 1 of 2 different and largely incompatible protocols
* the provision of one of several name-resolution services (chiefly, WINS or DNS)
* nodes must be told where to find the name server

If one also wants an outgoing connection, then one must also configure a gateway and so on.

Yes, TCP/IP works. I am not saying that it doesn't; that would be absurd. It is the universal protocol; everything speaks it now.

[geoffcampbell]

You are aware of autoconfiguration for IP? Works perfectly well for small networks, and gives exactly the plug-and-go capability you claim does not exist.

GJC

[liam_on_linux]

I've responded to that further down the page, in the hopes of not shattering the discussion even more.

[steer]

The reason network researchers talk in terms of the OSI stack is not because it was implemented (see "catastrophe of the two elephants") but because it places a protocol in context. The context of NetBeui as I understand it is that it as a packet shifting protocol it is layer 3 with no layer 4. That is, no reliability layer. May be wrong there. This means that it is simple to set up but your application layer programmer hates you. Might be wrong on that one though.

It also doesn't scale at all.

[geoffcampbell]

That seems like a fair summary to me. If I recall correctly (and this is all ancient history now, and I don't have Liam's obsessive near-eidetic memory) NETBIOS and Named Pipes were layered on top of NETBeui in a rather kludgy fashion.

It sort of worked, but it had none of the elegance and simplicity of TCP.

GJC

[liam_on_linux]

Interesting. That almost looks like a concession -- that you can't remember some of this stuff.

[geoffcampbell]

I can't remember the details, no. Why would I?

GJC

[liam_on_linux]

No, it didn't scale.

But again, let me spell this out:

I am not saying anything as palpably ridiculous as "[insert name of 1980s protocol] was better than TCP/IP."

What I am saying is much simpler and less contentious (I thought!). It is, approximately:

"1980s network protocols required less configuration than modern ones do."

What is "the catastrophe of 2 elephants"? I've never heard of it and neither has Google.

[steer]

http://students.depaul.edu/~jabsher/apoc_eleph/apoc_eleph.html

That's because I should have said apocalypse of two elephants: Essentially the place where the successful protocol emerges is somewhere between the "carefully thought out research idea based on solid mathematics and optimal behaviour" and "thing that people wanted to get to work and needed deployed in a hurry".

[liam_on_linux]

When I started out, everything spoke different protocols and to connect a machine to servers from different vendors required running multiple different protocols and client stacks, side-by-side, all talking to the same network card. In the days of DOS this was quite extraordinarily tricky to do, and if you actually succeeded, given that DOS only handled 640kiB of RAM, the chances are that you wouldn't have enough RAM left to run the apps you wanted.

At least Windows for Workgroups made it /relatively/ easy. That was significant progress.

geoffcampbell and I - although it may not look like it, we're old mates, honestly - have compared notes on the horrors of getting WfWg to talk to DECnet before now, something we both fought. His remarkably courageous approach freed up even more RAM than mine, which I had not thought possible.

Win95 made it actually easy, although a now-forgotten limitation was no more than 4 IP addresses per machine, which has bitten me before. Win98 made that go away and since then it's been plain sailing.

So, yes, TCP/IP, it's great, it's wonderful.

However, it dates to the 1970s, by the mid-1980s it was widespread and by the end of them it was becoming obvious that it was going to be The Protocol Of The Future.

I am amazed that in all those years, that by the time it got to microcomputers and PCs, its installation and configuration had not been simplified to the same level as NetBEUI or IPX/SPX or AppleTalk or DECnet. Where it was basically a case of turn it on and use it.

It still requires all that config to be done by someone, even today. It's still complex.

I had hoped that after over 4 *decades* of engineering effort and R&D that it would have been polished and simplified to rival the level of ease-of-use of its simpler competitors from *a quarter of a century ago*. And yet it hasn't happened.

We are, in terms of the number of things that have to be done by sysadmins, and done right, to prevent serious problems, worse off than we were 20y ago.

Yes, everyone's used to it. That doesn't mean it's OK.

[liam_on_linux]

Once upon a time, when you attached a disk to a computer, you had to know its filesystem, ensure that your OS either supported it already, or had the driver enabled, or possibly that appropriate 3rd-party software was installed, you had to have set its ID to something unique using jumpers or DIP switches, possibly configure its bus termination, and finally powered-down the machine, attached the drive, turned the drive on, then and only then turned the computer on, and crossed your fingers that you'd done everything right and the machine would boot and your drive would at least be readable. Probably not writable but you could at least see it and its contents. Then came the issue of reading the file formats, of course.

Now, you plug it in and it works. Live, online, more or less any disk from any machine into any other machine. Microsoft is intransigent - you can't read Mac or Linux disks on Windows. Everyone else supports everything.

Frankly, I had expected that by now, networks would be the same. That some godlike guru at a NOC somewhere set the magic parameters, or more to the point, ran a machine that did so for him, and every other computer on the Internet could just be plugged in and would just work. Automagically.

It hasn't happened.

I feel that this is a failure. A failure of ambition, of design and of implementation. Yes I know it's complex and hard. So is most of computing. But most of it has got easier since 1988. Networking has got harder.

Yes, I know, now we have everything connected to everything else, all the time. The security implications are massive.

But, still, given all the advancement, the fact that it is not easier, it is harder - that is /ridiculous./

[geoffcampbell]

If you really think any of that is harder now than it was in 1988, there really is no point in continuing this discussion. Get out of IT, you are entirely unsuited to it.

If you want whatever it is you are whining on about, stay in IT and design it. Don't whine that everyone else has made it too hard for you, show us how to do it better.

Shit or get off the pot, as my dear old Dad would have said to you.

GJC

[liam_on_linux]

Yes, it really is, and you have completely failed if you think that you have persuaded me otherwise.

I am not saying it does not work. I am not saying it is unacceptable. I am not denying that there are reasons for it. I am not saying it's a broken POS.

What I am /asking/ is this:

[1] has nobody else noticed this?
[2] wasn't it easier before?
[3] WIBNIF if was that easy today?

And *you* are not answering, you're merely shouting LALALA YOU'RE STUPID AND YOU SMELL, IT IS ALL FINE, THERE IS NOTHING AT ALL WRONG HERE, NOTHING TO SEE, HE'S MAKING IT ALL UP.