Major asia pacific public exchange points

John Milburn jem at
Mon Nov 1 03:09:21 UTC 1999

Sorry to sound nasty in some of the following, but every answer to
date has been so wrong, or wrong-headed.  Note that I am not in any
way unbiased here.  I built and operate L2IX, mentioned below, and
operate the largest ISP (2 million dialup, 8000 leased lines, 85% of
colo content, US $200 million 1999 revenue) in Korea (and, perhaps,
the largest in Asia-Pacific).

Of course any exchange in Korea will be primarily a "domestic" exchange.
WTF else could it be?  That is where the traffic is. Since there is
such a huge disparity between the cost of domestic connectivity and
international connectivity, it makes sense for domestic players to
exchange traffic amongst their domestic networks. Also, since Korea
is a unique language country, the overwhelming majority of the traffic
is domestic.  Only about 10% of my traffic is exchanged with the US.

There are two foreign ISPs active in Korea, PSINet (which bought the
Korean ISP INET last year) and AUNet. PSINet has its own international
connectivity, and AUNet buys theirs from Korea domestic resellers. Neither
of those players peers their international networks in Seoul, AUNet
because it is impossible given their connectivity (they have no direct
connectivity to other AUNet sites in Asia), and PSINet due to their US
centric policy.

Korea Telecom has their own domestic exchange service, but it operates
at Layer 3 only, and any exchange between ISPs sees the KT AS hop in the
path. This link is used by domestic ISPs primarily for connection to KT,
not to each other.

The only layer 2 exchange points in Korea are KINX and L2IX. KINX was
formed by an assiciation of smaller ISPs, led by PSINet. The really
large players (Dacom and KT) do not participate. The non commercial
networks also have restrictions on participating. The exchange is located
withing the computing center of the National Computerization Agency, and
is operated by NCA. This, along with some of the association policies,
creates concerns about the scalibility of the exchange. Current total
traffic is ~400 Mbps.

Dacom's L2IX is the largest (highest number of particiapnts, largest
traffic exchanged) NAP, with most Korean ISPs participating. The
total traffic throughput is now more than 2.0 Gbps. While KT does
not participate directly, Dacom provides domestic transit to KT (via
445 Mbps direct private peering), and the non-commercial networks,
at the exchange. The exchange is located in Dacom's main facility,
but early next year will move to the Korea Internet Data Center (KIDC),
a new 270,000 ft^2 (the size is not a typo) carrier neutral colocation
facility which we have created in Seoul.

International players are entirely welcome at KIDC/L2IX. Peering policies
are entirely bilateal matters, left to direct negotiation between the
participants. To peer with Dacom at this exchange, I require that the
international participant peer its entire network, as I do.

Now, saying that JPIX, NSPIXP2, STIX or HKIX are "international"
exchanges, and KINX or L2IX are not, is misleading. STIX is international
only in the sense that Singapore Telecom transits its international
peering connections to the Singapore domestic ISPs which connect to STIX.
JPIX and NSPIXP2 have some non-Japan participants, but they in general
peer only their Japan domestic traffic. HKIX, as an academically
controlled exchange, has serious scaling issues, and a multilateral
peering requirement which is causing increasing trouble. Though UUNET
participates there, what traffic do they exchange? Certainly not 701.
How about their OzEmail subsidiary?

One needs to be very, very careful when characterizing the nature of
exchanges, since there are many different parameters of operation for
those who operate outside of the US, and currently pay the full cost of
connecting to the US. Peering policies at thesed exchanges might have
a significant effect on US traffic, particularly in English-centric,
small markets like Singapore and Hong Kong, where the majority of the
traffic is international.

US based Internet players, in general, pay nothing for international
connectivity, and are very cautious to not change this, even when they
do participate in some low level at an "international" exchange. There
are only a couple of examples of US ISPs (AboveNet and Concentric, that I
know of), which peer their entire network at some exchange, mostly LINX.
Most others (who even have an international presence) separate their
networks in US, Europe, and Asia portions, and many add country specific
breakdown. Only in the US do they peer the entire network. Elsewhere,
it is always a subset, with the local ISP still paying the costs of
international trasit to the US customers of the big ISP. IMNSHO, these
local ISPs are fools to peer with big US ISPs under these terms.

Additionally, local exchanges tend to be loaded with local competitive
issues.  Local ISPs see their major competitors as other local ISPs,
and thus use peering policy and capacity to engineer competitive and/or
political policy. This should be more than familiar to all of the US
based readers of NANOG.


     John Milburn             jem at, jem at
     Managing Director      Internet Technology Division
     Cell +82 19-220-7035            Tel +82 22-220-7035
     Dacom Corporation, Seoul, Korea Fax +82 22-220-7429

bmanning at writes:
>>  :| I think there's also one in Seoul called KIX as well, but I'm not entirely
>>  :| sure.
>> Yes, KIX is in Seoul but I think it is domestic IX.  In 
>> Japan as well there is no IXes which can be called 
>> international one.
>	As I understand, KIX is mostly domestic Korea.
>	Japan has NSPIX3 and JPIX.  There is enough
>	international capacity to house a root server
>	there. :)
>	And based on what I have been told, both HKIX and 
>	STIX can be called international.  There is a 
>	common error in assuming international = US domestic.
>	This is false.
>	Both STIX & HKIX have interconnections to ISPs in
>	many other nations but not quite the connectivity back
>	to the US.

More information about the NANOG mailing list