Re: Just an idea for a self organiziong IPv6 address space network

Jing Shen <[email protected]> Thu, 7 Nov 2002 11:11:39 +0800 (CST)
Newsgroups gmane.network.research.irtf
Message-ID <[email protected]>
--0-311075602-1036638699=:58053
Content-Type: text/plain; charset=gb2312
Content-Transfer-Encoding: quoted-printable
X-MIME-Autoconverted: from 8bit to quoted-printable by puck.nether.net id gA73Cx9H025100


 Hi all,
the situation Yang thought about is the situation some ISPs do with their=
 customers who are=20
small companies. These companies just want to access the internet with ac=
ceptable access speed=20
but with limited budget. So, they just buy some IP address from ISP and l=
ease line with low speed ( as I know
it's usually under 1Mbps). But the situation changes when ADSL and LAN ac=
cess come into market. By using monitoring
tools, ADSL could be kept up nearly all the time , and by using  NAT and =
ADSL (or 10Mbps ethernet LAN), small companies
 can easily set up their internet connection, and with help of DynaDNS th=
ey can even set up web service, email service etc.=20
So, more and more small companies abandon the old way of paying for a blo=
ck of IP address in ISP network.=20
The second,  from the point of view of routing, AS enables  a reliable an=
d dynamic routing ability in internet. The path vector
is exchanged and stored in routers  while only the 'best' one is used. Th=
is means, no matter a company's network
is single homed or multi-homed it can rely on the BGP to get a fulltime c=
onnection, with help of policy based routing some commercial
policy could be applied too.  Within AS,  OSPF/IS-IS could be used to est=
ablish best paths.=20
To the extrem situation, if a customer hire two leased line from two ISP,=
 and each ISP allocate a block of ip address to the customer,=20
policy based routing and NAT could guarantee the internet access but a bl=
ock of IP address is wasted.  I think this situation is the one=20
considered by Yang, and maybe he's thinking about a solution by routing. =
But  I think this could solved by using solution using
(DHCP, network monitoring, two or more IP block from two AS, two or more =
leased line). In this solution, a DHCP server hold all=20
addresses and monitors outgoing lines, when a computer comes up the DHCP =
server allocate ip address according to state of leased=20
lines.  The access router just need to set up static routing table ( mayb=
e NAT).  The problem exists with those servers, who could not=20
change its ip address from time to time and the cost.=20
So,  to my opinion it's not a good idea to allocate IP addres hierachical=
ly and a peer connection between AS is a good solution.
regards
  Xiaowei Yang <[email protected]> =B5=C4=D5=FD=CE=C4=A3=BA Thanks=
 for the explanations and for the pointers. I agree that
dynamically allocating IP addresses is not desirable in practice, but
I do think it is possible to "statically" allocate provider-rooted
hierarchical addresses. (I was not the one originating the thread, but
my thesis work is related to the topic.)

What I mean by "statically" is that once a provider-customer
relationship is established, the customer is allocated a fixed address
block from the provider. A multi-homed customer will be allocated one
or more address blocks from each of its providers.

I agree that there might be reluctance to adopt this approach. This
is a general problem discussed in Shenker et al's paper on algorithmic
mechanism design. In a distributed system like Internet, we can not
simply expect that each AS cares about the stability and scalability
of BGP and will follow the protocols we design. Each AS will act
selfishly, but we can design protocols with incentives for an AS to
follow.

As an example, if an global address prefix costs a significant amount
of fortune, perhaps many small networks / local ISPs are willing to
take address blocks from their providers. If becoming a visible entry
in the global BGP table costs money, perhaps many networks would want
to be aggregated into their providers' entries.

By this allocation, ISPs can still operate if the rest of the Internet
collapses, because dynamic failures do not terminate the business
relationships and do not change the address blocks a domain owns.

We do not need algorithms to determine which ASes are the top level
ones. Whichever affords the money can buy a global unique address
prefix.

Due to the hierarchical aggregation, some dynamic failure will be
hidden from outside the address blocks. Thus, the end system itself
will need to have some error recovery mechanisms. Once an address
fails, it tries another one.

Kleinrock's hierarchical routing scheme cannot be directly applied to
the Internet. In Kleinrock's work, there is no "policy routing"
involved. Each node is willing to provide transit service for each
other. Thus, the cluster boundaries can be formed easily. In Internet,
due to policy routing, there is no natural way of forming clusters.

However, landmark routing can be applied to the Internet. This
hierarchical allocation scheme can be viewed as a variant of landmark
routing. A provider is the landmark for a customer. Packets are routed
towards the landmarks unless a better route is heard.




Jing Shen

State Key Lab of CAD&CG
ZheJiang University(YuQuan)
HangZhou, ZheJiang Province 310027
P.R.China


---------------------------------
Do You Yahoo!?
"=CA=C7IT=BE=AB=D3=A2=C2=F0=A3=BF=D0=A1=CA=D4=C5=A3=B5=B6=BB=F1=CA=B1=C9=D0=
=B4=F3=BD=B1=A3=A1"
--0-311075602-1036638699=:58053
Content-Type: text/html; charset=gb2312
Content-Transfer-Encoding: quoted-printable
X-MIME-Autoconverted: from 8bit to quoted-printable by puck.nether.net id gA73Cx9H025100

<P>&nbsp;Hi all,
<P>the situation Yang thought about is the situation some ISPs do with th=
eir customers who&nbsp;are=20
<P>small companies. These companies just want to access the internet with=
 acceptable access speed=20
<P>but with limited budget. So, they just buy some IP address from ISP an=
d lease line with low speed ( as I know
<P>it's usually under 1Mbps). But the situation changes when ADSL and LAN=
 access come into market. By using monitoring
<P>tools, ADSL could be kept up nearly all the time , and by using&nbsp; =
NAT and ADSL (or 10Mbps ethernet LAN), small companies
<P>&nbsp;can easily set up their internet connection, and with help of Dy=
naDNS they can even set up web service, email service etc.=20
<P>So, more and more small companies abandon the old way of paying for a =
block of IP address in ISP network.=20
<P>The second,&nbsp; from the point of view of routing, AS enables&nbsp; =
a reliable and dynamic routing ability in internet. The path vector
<P>is exchanged and stored in routers &nbsp;while only the 'best' one is =
used. This means, no matter a company's network
<P>is single homed or multi-homed it can rely on the BGP to get a fulltim=
e connection, with help of policy based routing some commercial
<P>policy could be applied too.&nbsp; Within AS,&nbsp; OSPF/IS-IS could b=
e used to establish best paths.&nbsp;
<P>To the extrem situation, if a customer hire two leased line from two I=
SP, and each ISP allocate a block of ip address to the customer,=20
<P>policy based routing and NAT could guarantee the internet access but a=
 block of IP address is wasted.&nbsp; I think this situation is the one=20
<P>considered by Yang, and maybe he's thinking about a solution by routin=
g. But&nbsp; I think this could solved by using solution using
<P>(DHCP, network monitoring, two or more IP block from two AS, two or mo=
re leased line). In this solution, a DHCP server hold all=20
<P>addresses and monitors outgoing lines, when a computer comes up the DH=
CP server allocate ip address according to state of leased=20
<P>lines.&nbsp; The access router just need to set up static routing tabl=
e ( maybe NAT).&nbsp; The problem exists with those servers, who could no=
t=20
<P>change its ip address from time to time and the cost.=20
<P>So,&nbsp; to my opinion it's not a good idea to allocate IP addres hie=
rachically and a peer connection between AS is a good solution.
<P>regards
<P>&nbsp; <B>Xiaowei Yang &lt;[email protected]&gt;</B> =B5=C4=D5=FD=
=CE=C4=A3=BA=20
<BLOCKQUOTE style=3D"PADDING-LEFT: 5px; MARGIN-LEFT: 5px; BORDER-LEFT: #1=
010ff 2px solid">Thanks for the explanations and for the pointers. I agre=
e that<BR>dynamically allocating IP addresses is not desirable in practic=
e, but<BR>I do think it is possible to "statically" allocate provider-roo=
ted<BR>hierarchical addresses. (I was not the one originating the thread,=
 but<BR>my thesis work is related to the topic.)<BR><BR>What I mean by "s=
tatically" is that once a provider-customer<BR>relationship is establishe=
d, the customer is allocated a fixed address<BR>block from the provider. =
A multi-homed customer will be allocated one<BR>or more address blocks fr=
om each of its providers.<BR><BR>I agree that there might be reluctance t=
o adopt this approach. This<BR>is a general problem discussed in Shenker =
et al's paper on algorithmic<BR>mechanism design. In a distributed system=
 like Internet, we can not<BR>simply expect that each AS cares about the =
stability and scalability<BR>of BGP and will follow the protocols we desi=
gn. Each AS will act<BR>selfishly, but we can design protocols with incen=
tives for an AS to<BR>follow.<BR><BR>As an example, if an global address =
prefix costs a significant amount<BR>of fortune, perhaps many small netwo=
rks / local ISPs are willing to<BR>take address blocks from their provide=
rs. If becoming a visible entry<BR>in the global BGP table costs money, p=
erhaps many networks would want<BR>to be aggregated into their providers'=
 entries.<BR><BR>By this allocation, ISPs can still operate if the rest o=
f the Internet<BR>collapses, because dynamic failures do not terminate th=
e business<BR>relationships and do not change the address blocks a domain=
 owns.<BR><BR>We do not need algorithms to determine which ASes are the t=
op level<BR>ones. Whichever affords the money can buy a global unique add=
ress<BR>prefix.<BR><BR>Due to the hierarchical aggregation, some dynamic =
failure will be<BR>hidden from outside the address blocks. Thus, the end =
system itself<BR>will need to have some error recovery mechanisms. Once a=
n address<BR>fails, it tries another one.<BR><BR>Kleinrock's hierarchical=
 routing scheme cannot be directly applied to<BR>the Internet. In Kleinro=
ck's work, there is no "policy routing"<BR>involved. Each node is willing=
 to provide transit service for each<BR>other. Thus, the cluster boundari=
es can be formed easily. In Internet,<BR>due to policy routing, there is =
no natural way of forming clusters.<BR><BR>However, landmark routing can =
be applied to the Internet. This<BR>hierarchical allocation scheme can be=
 viewed as a variant of landmark<BR>routing. A provider is the landmark f=
or a customer. Packets are routed<BR>towards the landmarks unless a bette=
r route is heard.<BR><BR><BR></BLOCKQUOTE><BR><BR>Jing Shen<br><br>State =
Key Lab of CAD&amp;CG<br>ZheJiang University(YuQuan)<br>HangZhou, ZheJian=
g Province 310027<br>P.R.China<p><br><hr size=3D1><b>Do You Yahoo!?</b><b=
r>
<a href=3D"http://rd.yahoo.com/mail_cn/tag/?http://cn.promo.yahoo.com/cgi=
-bin/udb/u">"=CA=C7IT=BE=AB=D3=A2=C2=F0=A3=BF=D0=A1=CA=D4=C5=A3=B5=B6=BB=F1=
=CA=B1=C9=D0=B4=F3=BD=B1=A3=A1"</a>
--0-311075602-1036638699=:58053--
_______________________________________________
irtf-rr mailing list
[email protected]
http://puck.nether.net/mailman/listinfo/irtf-rr