I agree that most companies will not want true end-to-end Internet
connectivity in their networks, and that there is no real way to create the
connectivity when it is intentionally prevented (nor should you try)  I was
speaking only of the case where the connectivity is interrupted as a side
effect of using NAT (hence my comment 'I assume you mean ... when it is not
intentionally disrupted by a firewall').  However, I should have been more
clear that IPv6/Teredo is no where near a silver bullet for P2P
communication - thanks for making that explicit.

Its true that a NAT device has some of the properties of a basic firewall
(prevents arbitrary inbound connections), but that is not what NAT is
designed for.  Sure many people have come to see the effects of NAT as a
form of rudimentary security, but any organization that is relying on a
plain NAT device to protect their network instead of using a firewall is
making a serious mistake.  If you want to limit the traffic into and out of
your network, use a firewall - that's exactly what its designed for and that
is the only way you'll have real controls over the traffic.  Besides, any
commercial quality NAT device will have full blown firewall support built in
as well.

Because NATs aren't designed as a security mechanism, they have several
limitations when used in place of a firewall.  For example, the most basic
firewall rule like 'prevent all incoming packets except for those involved
in TCP sessions that have been initiated from inside my network' (an
extremely common first rule) is much stronger than the protection offered by
most NATs.  In the most common type of NAT device (a 'cone NAT') if a host
'A' inside your network sends a packet from port 'a' to an external host
'B', and unrelated host on the internet 'C' can send arbitrary packets back
to 'A' port 'a'.  In fact, this is the exact mechanism that Teredo uses to
tunnel IPv6 traffic over a NAT.  NATs generally also don't make any attempt
to prevent IP spoofing, or other network attacks.

Sure there are probably a lot of small (and maybe even some big)
organizations relying on a simple NAT device to protect their network, but
as you said the issue isn't really a technology problem but an IT/management
one.  In reality, if you can't control the software that's running on your
machines inside your network, then there is really nothing you can do to
really protect your network, short of cutting of all Internet access.  After
all, even if you only give me the ability to browse web-pages via a secured
http proxy, I can always build a IP-over-HTTP tunnel that would allow
arbitrary hosts on the internet to tunnel arbitrary traffic to any host
inside the 'secured' network.  Microsoft acknowledges the potential problem
of organizations having their inadequate firewalls circumvented by tunneling
technology like Teredo by disabling Teredo on any machine that is a member
of a domain (in the P2P beta, you couldn't manually enable this, in the
final IPv6 release I believe you can).

The other benefit of NAT that you mentioned is that it hides information
(presumably the number and identity of hosts) about your network from the
outside world.  This is certainly true, and I agree that some organizations
who are already using a firewall to prevent inbound connections may continue
to use NAT for this purpose. Then again, these organizations may be more
likely to just require all traffic pass through a proxy server that can also
try to prevent other information leakage (eg. HTTP cookies).  However, I
believe there are many situations where organizations would be happy with a
more open network (eg. inbound connections permitted to specific ports for
specific applications like IM or other group collaboration) but are using
NAT because of the difficulty and hassle involved in getting and maintaining
routable IP addresses for all of their workstations.

Typically, in order to get a larger IP address block from your upstream ISP
today you have to prove that you're making adequate use of the majority of
your current block, and then go through a potentially painful and slow
renumbering process just to get double the address size (not to mention
probably pay more).  At the rate at which some organizations grow, there is
a real danger that you won't be able to convince your ISP to expand your IP
range fast enough to accommodate new nodes (I had this problem when growing
the modem pool at an ISP I worked for many years ago).  Today its just
easier and safer to use NAT in almost all network situations (I haven't even
mentioned home networks where getting more than one IP address is often
impossible or very expensive).  In the IPv6 world, the smallest address
block you can get for a network (eg. even a home dialup user) supports up to
2^64 nodes!  In this world, its a lot easier to NOT use NAT, especially when
you consider NAT will break some of the capabilities of IPv6 (eg. end-to-end
encryption with IPsec) and incurs more overhead than simple routing (payload
checksums have to be recomputed).

Anyway, personally I'd love to see IPv6 replace IPv4, and NAT replaced with
better technology targeted to the problem being solved.  Practically, its
anyone's guess as to what will ultimately happen (although the fact that
this biggest OS and software provider in the world is currently banking on
IPv6 is promising).  Given the slowness of change and lack of deep
understanding in IT departments (have you ever tried to convince a network
admin their routing config is creating a PMTUd black hole?), IPv6 may never
be universally adopted and we might live perpetually in a world of repeated
address translation and proxying (NAT's expansion potential is not
unlimited, its bound by the 16-bit port-space which is a very real limit in
large networks).  One thing is clear though, there is an increasing demand
for P2P applications, even in (perhaps especially in) a business setting,
which is making an increasing demand for improved Internet connectivity.  If
this trend continues, SOMETHING has to change in the currently loosely
connected IPv4/NAT world.

Rick

----- Original Message -----
From: "Griffiths, Ian" <[EMAIL PROTECTED]>
To: <[EMAIL PROTECTED]>
Sent: Thursday, September 04, 2003 12:56 PM
Subject: Re: .NET P2P Network Programming


> One thing that's also worth pointing out is that there are many firms
> which consider the fact their IPv4 firewalls don't allow incoming
> connections to be a feature, not a limitation.
>
> NAT serves more than one purpose.  It seems that you are focusing on the
> fact that it provides a short terms solution to the shortage of IP
> addresses.  But it also provides the useful facility of hiding network
> structure behind a single facade - it makes it much more difficult to
> infer things about a company's network structure from outside.  From
> what you're saying, it seems like you're regarding this as an
> unfortunate side effect of the way that NAT works, but actually a lot of
> organizations regard this as being the main reason for using it, with
> the reduced need for IP addresses being merely a bonus.
>
> So it is dangerous to assume that the problems you are coming up against
> will necessarily be solved even when IPv6 is more or less ubiquitous
> (however many years off that may be).  I would expect many (and maybe
> even most) companies to carry on using some form of NAT precisely
> because they *want* all of their externally visible traffic to look like
> it comes from one address in order to minimize the amount of information
> about their internal networks that leaks out to the outside world.
>
> And likewise, just because IPv6 provides a means of allowing end to end
> p2p connections, you shouldn't assume that you'll necessarily be able to
> exploit this.  Consider the situation we have today even where NAT is
> not in use - there are plenty of hosts on the internet which aren't
> behind NAT, but even these are often not able to open up ports for p2p
> connections despite having routable addresses.  This is because such
> machines are very often behind packet filters or firewalls.
>
> In most companies, the IS department will have set up their network so
> that all incoming requests are dropped apart from the small fixed set of
> services they have chosen to support.  (E.g. the http, and https ports
> will be open for a web server.  The smtp, pop3, and imap ports will be
> open for a mail server.  Everything else will be shut.)  This is basic
> good practice, and the people who don't follow this are the same people
> who got hit by things like MSBLAST and Slammer.
>
> Try convincing a bank's IS department that you'd like your desktop to be
> able to accept incoming connections.  You are pretty unlikely to get as
> far as a discussion of the relative merits of UPnP vs. IPv6 - you're
> most likely just to be told "No".
>
> There's a reason UPnP isn't universally available: lack of demand.  If
> everyone responsible for administrating NAT firewalls wanted the ability
> to run p2p software behind the firewall, you can be sure that UPnP or
> some equivalent 'hack' would be universally available.  But most network
> administrators in commercial organizations positively don't want this
> (and many would regard the whole point of having a firewall being to
> prevent this kind of thing).  So don't expect life to get better even
> when IPv6 is universal - filtering will be widely deployed just as it is
> today in routable IPv4 scenarios.
>
> What it boils down to is that this isn't really a technical issue.  It
> looks like one on the face of it, but it's really a socio-economic one:
> a large proportion of people on the internet can't do p2p because their
> bosses don't want them to.  Technological solutions aren't going to
> change that.  (Compelling applications might.  The web was compelling
> enough to convince a vast number of companies to connect themselves up
> to a public and fundamentally unsecure network...)
>
>
> Of course, as long as you realize this, then that's no reason not to use
> IPv6.  It sounds like an interesting plan, an from a technical
> perspective it's more aesthetically pleasing than the IP4 alternatives.
> It's just important not to overestimate the scope that an IPv6 solution
> will offer you - your application will still be subject to the same
> limitations as an IPv4 one in the real world because ultimately, those
> limitations are primarily not technological.
>
> I'm just worried about what you said here:
>
> > My main reason for wanting to use IPv6 cause as you well stated
> > the IPv4 solutions are merely hacks that cannot guarantee a working
> > solution due to different scenarios -
>
> This implies that you think that IPv6 *can* guarantee a working
> solution.  It cannot.  The fundamental problem is that most corporations
> regard the limitations imposed by NAT firewalls as a positive feature,
> not a limitation.
>
>
> --
> Ian Griffiths
> DevelopMentor
>
>
> > -----Original Message-----
> > From: Mark Anthony Spiteri [mailto:[EMAIL PROTECTED]
> >
> > Thanks a lot Rick for you reply.
> >
> > Yes in fact that's why I was looking into using IPv6 as Microsoft
> > suggests.
> > My main reason for wanting to use IPv6 cause as you well stated the
> IPv4
> > solutions are merely hacks that cannot guarantee a working solution
> due to
> > different scenarios - especially the fact that UPnP is not supported
> by
> > all
> > hardware as yet.
> >
> > The issue you mentioned about the OS limitation is not particularly an
> > issue
> > since this is a University degree project. Using IPv6 will be a great
> > benefit in terms of academic content.
> >
> > Guess the only issue then is to learn how to use netsh :)
> >
> > Cheers,
> > Mark
> >
> > -----Original Message-----
> > From: Rick Byers
> >
> > By "the NAT problem", I assume you mean restoring end-to-end
> > connectivity on the Internet (when it is not intentionally disrupted
> > by a firewall).
> > There are all sorts of ways of doing this with IPv4 ("bi-directional'
> > connections, UPnP, etc) but in my opinion they're all just hacks
> > that greatly increase the complexity of the problem and only work
> > around specific cases of the problem (eg. with 'bi-directional'
> > channels, one end has to be directly connected to the Internet).
> >
> > Microsoft is taking a longer term approach by encouraging the use
> > of IPv6 which has an expanded address space and makes NAT unnecessary.
> >  In fact, MS's P2P SDK (http://www.microsoft.com/windowsxp/p2p/)
> > requires IPv6. In an ideal world, everyone would just upgrade their
> > routers to support IPv6 and we could just forget about NATs.
> > Unfortunately this is an excruciatingly slow transition, its even
> > taken Microsoft many years to release an os (Windows Server 2003)
> > with a production quality IPv6 stack, other OSes like
> > NetBSD have had one since before 1999.  So a lot of effort is required
> > to tunnel IPv6 traffic over the existing IPv4 internet, including
> > dealing with IPv4 NATs.
> >
> > First of all, I think encouraging the use of IPv6 is a great
> > approach because its the only way we'll ever make the transition,
> > I'm extremely happy with Microsoft's position (there is no
> > technical reason their P2P stuff couldn't support IPv4 as well as
> > IPv6).  Microsoft's Teredo technology
> >
> (http://www.microsoft.com/technet/prodtechnol/winxppro/maintain/Teredo.a
> sp
> > )
> > is designed to tunnel IPv6 traffic across an IPv4 NAT.  Like the
> > other IPv4 NAT technologies, its still a hack that we'd be much better
> > off without.
> > However, in my opinion, its a much more reliable, efficient and
> > elegant solution to NAT traversal than any of the other choices.
> >  Its interesting to note that there is nothing special about IPv6
> > that makes teredo NAT traversal possible, you could implement
> > something similar using IPv4 (although I'm very happy Microsoft
> hasn't).
> >
> > Anyway, enough ranting background material and on to your question.
> > Teredo is an OS-level technology, and there is nothing you have to
> > do to make use of it from within an application.  So, yes, using
> > .NET (v1.1 and up) you can write IPv6 networking code (or better yet,
> > IP version agnostic networking code) and rely on the underlying OS
> > networking layer to take care of all the NAT traversal for you.  Of
> > course you'll have to learn to configure, manage, and possibly debug
> > the IPv6 support in Windows (If you've never used 'netsh'
> > before, its time to learn).  The biggest problem with this approach
> > is OS support.  Windows Server 2003 is the only MS OS that supports
> > IPv6 natively, but the stack is available as a 'recommended upgrade'
> > for Windows XP.  For anything older than XP, you're pretty much out
> > of luck.
>
> ===================================
> This list is hosted by DevelopMentor�  http://www.develop.com
> NEW! ASP.NET courses you may be interested in:
>
> 2 Days of ASP.NET, 29 Sept 2003, in Redmond
> http://www.develop.com/courses/2daspdotnet
>
> Guerrilla ASP.NET, 13 Oct 2003, in Boston
> http://www.develop.com/courses/gaspdotnet
>
> View archives and manage your subscription(s) at
http://discuss.develop.com
>

===================================
This list is hosted by DevelopMentor�  http://www.develop.com
NEW! ASP.NET courses you may be interested in:

2 Days of ASP.NET, 29 Sept 2003, in Redmond
http://www.develop.com/courses/2daspdotnet

Guerrilla ASP.NET, 13 Oct 2003, in Boston
http://www.develop.com/courses/gaspdotnet

View archives and manage your subscription(s) at http://discuss.develop.com

Reply via email to