Independent Submission J. Wu
Request for Comments: 5747 Y. Cui
Category: Experimental X. Li
ISSN: 2070-1721 M. Xu
Tsinghua University
C. Metz
Cisco Systems, Inc.
March 2010
4over6 Transit Solution Using IP Encapsulation and MP-BGP Extensions
Abstract
The emerging and growing deployment of IPv6 networks will introduce
cases where connectivity with IPv4 networks crossing IPv6 transit
backbones is desired. This document describes a mechanism for
automatic discovery and creation of IPv4-over-IPv6 tunnels via
extensions to multiprotocol BGP. It is targeted at connecting
islands of IPv4 networks across an IPv6-only backbone without the
need for a manually configured overlay of tunnels. The mechanisms
described in this document have been implemented, tested, and
deployed on the large research IPv6 network in China.
Status of This Memo
This document is not an Internet Standards Track specification; it is
published for examination, experimental implementation, and
evaluation.
This document defines an Experimental Protocol for the Internet
community. This is a contribution to the RFC Series, independently
of any other RFC stream. The RFC Editor has chosen to publish this
document at its discretion and makes no statement about its value for
implementation or deployment. Documents approved for publication by
the RFC Editor are not a candidate for any level of Internet
Standard; see Section 2 of RFC 5741.
Information about the current status of this document, any errata,
and how to provide feedback on it may be obtained at
http://www.rfc-editor.org/info/rfc5747.
IESG Note
The mechanisms and techniques described in this document are related
to specifications developed by the IETF softwire working group and
published as Standards Track documents by the IETF, but the
relationship does not prevent publication of this document.
Wu, et al. Experimental [Page 1]
RFC 5747 4over6 March 2010
Copyright Notice
Copyright (c) 2010 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(
http://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect
to this document.
Table of Contents
1. Introduction ....................................................3
2. 4over6 Framework Overview .......................................3
3. Prototype Implementation ........................................5
3.1. 4over6 Packet Forwarding ...................................5
3.2. Encapsulation Table ........................................6
3.3. MP-BGP 4over6 Protocol Extensions ..........................7
3.3.1. Receiving Routing Information from Local CE .........8
3.3.2. Receiving 4over6 Routing Information from a
Remote 4over6 PE ....................................8
4. 4over6 Deployment Experience ....................................9
4.1. CNGI-CERNET2 ...............................................9
4.2. 4over6 Testbed on the CNGI-CERNET2 IPv6 Network ............9
4.3. Deployment Experiences ....................................10
5. Ongoing Experiment .............................................11
6. Relationship to Softwire Mesh Effort ...........................12
7. IANA Considerations ............................................12
8. Security Considerations ........................................13
9. Conclusion .....................................................13
10. Acknowledgements ..............................................13
11. Normative References ..........................................14
Wu, et al. Experimental [Page 2]
RFC 5747 4over6 March 2010
1. Introduction
Due to the lack of IPv4 address space, more and more IPv6 networks
have been deployed not only on edge networks but also on backbone
networks. However, there are still a large number of legacy IPv4
hosts and applications. As a result, IPv6 networks and IPv4
applications/hosts will have to coexist for a long period of time.
The emerging and growing deployment of IPv6 networks will introduce
cases where connectivity with IPv4 networks is desired. Some IPv6
backbones will need to offer transit services to attached IPv4 access
networks. The method to achieve this would be to encapsulate and
then transport the IPv4 payloads inside IPv6 tunnels spanning the
backbone. There are some IPv6/IPv4-related tunneling protocols and
mechanisms defined in the literature. But at the time that the
mechanism described in this document was introduced, most of these
existing techniques focused on the problem of IPv6 over IPv4, rather
than the case of IPv4 over IPv6. Encapsulation methods alone, such
as those defined in [RFC2473], require manual configuration in order
to operate. When a large number of tunnels are necessary, manual
configuration can become burdensome. To the above problem, this
document describes an approach, referred to as "4over6".
The 4over6 mechanism concerns two aspects: the control plane and the
data plane. The control plane needs to address the problem of how to
set up an IPv4-over-IPv6 tunnel in an automatic and scalable fashion
between a large number of edge routers. This document describes
experimental extensions to Multiprotocol Extension for BGP (MP-BGP)
[RFC4271] [RFC4760] employed to communicate tunnel endpoint
information and establish 4over6 tunnels between dual-stack Provider
Edge (PE) routers positioned at the edge of the IPv6 backbone
network. Once the 4over6 tunnel is in place, the data plane focuses
on the packet forwarding processes of encapsulation and
decapsulation.
2. 4over6 Framework Overview
In the topology shown in Figure 1, a number of IPv6-only P routers
compose a native IPv6 backbone. The PE routers are dual stack and
referred to as 4over6 PE routers. The IPv6 backbone acts as a
transit core to transport IPv4 packets across the IPv6 backbone.
This enables each of the IPv4 access islands to communicate with one
another via 4over6 tunnels spanning the IPv6 transit core.
Wu, et al. Experimental [Page 3]
RFC 5747 4over6 March 2010
_._._._._ _._._._._
| IPv4 | | IPv4 |
| access | | access |
| island | | island |
_._._._._ _._._._._
| |
Dual-Stack Dual-Stack
"4over6 PE" "4over6 PE"
| |
| |
__+____________________+__
4over6 / : : : : \ IPv6 only
Tunnels | : : : : | transit core
between | : [P] : | with multiple
PEs | : : : : | [P routers]
| : : : : |
\_._._._._._._._._._._._._./
| / \ |
| |
Dual-Stack Dual-Stack
"4over6 PE" "4over6 PE"
| | |
_._._._._ _._._._._
| IPv4 | | IPv4 |
| access | | access |
| island | | island |
_._._._._ _._._._._
Figure 1: IPv4 over IPv6 Network Topology
As shown in Figure 1, there are multiple dual-stack PE routers
connected to the IPv6 transit core. In order for the ingress 4over6
PE router to forward an IPv4 packet across the IPv6 backbone to the
correct egress 4over6 PE router, the ingress 4over6 PE router must
learn which IPv4 destination prefixes are reachable through each
egress 4over6 PE router. MP-BGP will be extended to distribute the
destination IPv4 prefix information between peering dual-stack PE
routers. Section 4 of this document presents the definition of the
4over6 protocol field in MP-BGP, and Section 5 describes MP-BGP's
extended behavior in support of this capability.
After the ingress 4over6 PE router learns the correct egress 4over6
PE router via MP-BGP, it will forward the packet across the IPv6
backbone using IP encapsulation. The egress 4over6 PE router will
receive the encapsulated packet, remove the IPv6 header, and then
forward the original IPv4 packet to its final IPv4 destination.
Section 6 describes the procedure of packet forwarding.
Wu, et al. Experimental [Page 4]
RFC 5747 4over6 March 2010
3. Prototype Implementation
An implementation of the 4over6 mechanisms described in this document
was developed, tested, and deployed on Linux with kernel version 2.4.
The prototype system is composed of three components: packet
forwarding, the encapsulation table, and MP-BGP extensions. The
packet forwarding and encapsulation table are Linux kernel modules,
and the MP-BGP extension was developed by extending Zebra routing
software.
The following sections will discuss these parts in detail.
3.1. 4over6 Packet Forwarding
Forwarding an IPv4 packet through the IPv6 transit core includes
three parts: encapsulation of the incoming IPv4 packet with the IPv6
tunnel header, transmission of the encapsulated packet over the IPv6
transit backbone, and decapsulation of the IPv6 header and forwarding
of the original IPv4 packet. Native IPv6 routing and forwarding are
employed in the backbone network since the P routers take the 4over6
tunneled packets as just native IPv6 packets. Therefore, 4over6
packet forwarding involves only the encapsulation process and the
decapsulation process, both of which are performed on the 4over6 PE
routers.
Tunnel from Ingress PE to Egress PE
---------------------------->
Tunnel Tunnel
Entry-Point Exit-Point
Node Node
+-+ IPv4 +--+ IPv6 Transit Core +--+ IPv4 +-+
|S|-->--//-->--|PE|=====>=====//=====>=====|PE|-->--//-->--|D|
+-+ +--+ +--+ +-+
Original Ingress PE Egress PE Original
Packet (Encapsulation) (Decapsulation) Packet
Source Destination
Node Node
Figure 2: Packet Forwarding along 4over6 Tunnel
As shown in Figure 2, packet encapsulation and decapsulation are both
on the dual-stack 4over6 PE routers. Figure 3 shows the format of
packet encapsulation and decapsulation.
Wu, et al. Experimental [Page 5]
RFC 5747 4over6 March 2010
+----------------------------------//-----+
| IPv4 Header | Packet Payload |
+----------------------------------//-----+
< Original IPv4 Packet >
|
|(Encapsulation on ingress PE)
|
v
< Tunnel IPv6 Headers > < Original IPv4 Packet >
+-----------+ - - - - - +-------------+-----------//--------------+
| IPv6 | IPv6 | IPv4 | |
| | Extension | | Packet Payload |
| Header | Headers | Header | |
+-----------+ - - - - - +-------------+-----------//--------------+
< Tunnel IPv6 Packet >
|
|(Decapsulation on egress PE)
|
v
+----------------------------------//-----+
| IPv4 Header | Packet Payload |
+----------------------------------//-----+
< Original IPv4 Packet >
Figure 3: Packet Encapsulation and Decapsulation on Dual-Stack 4over6
PE Router
The encapsulation format to apply is IPv4 encapsulated in IPv6, as
outlined in [RFC2473].
3.2. Encapsulation Table
Each 4over6 PE router maintains an encapsulation table as depicted in
Figure 4. Each entry in the encapsulation table consists of an IPv4
prefix and its corresponding IPv6 address. The IPv4 prefix is a
particular network located in an IPv4 access island network. The
IPv6 address is the 4over6 virtual interface (VIF) address of the
4over6 PE router that the IPv4 prefix is reachable through. The
encapsulation table is built and maintained using local configuration
information and MP-BGP advertisements received from remote 4over6 PE
routers.
The 4over6 VIF is an IPv6 /128 address that is locally configured on
each 4over6 router. This address, as an ordinary global IPv6
address, must also be injected into the IPv6 IGP so that it is
reachable across the IPv6 backbone.
Wu, et al. Experimental [Page 6]
RFC 5747 4over6 March 2010
+-------------+------------------------------------------------+
| IPv4 Prefix | IPv6 Advertising Address Family Border Router |
+-------------+------------------------------------------------+
Figure 4: Encapsulation Table
When an IPv4 packet arrives at the ingress 4over6 PE router, a lookup
in the local IPv4 routing table will result in a pointer to the local
encapsulation table entry with the matching destination IPv4 prefix.
There is a corresponding IPv6 address in the encapsulation table.
The IPv4 packet is encapsulated in an IPv6 header. The source
address in the IPv6 header is the IPv6 VIF address of the local
4over6 PE router and the destination address is the IPv6 VIF address
of the remote 4over6 PE router contained in the local encapsulation
table. The packet is then subjected to normal IPv6 forwarding for
transport across the IPv6 backbone.
When the encapsulated packet arrives at the egress 4over6 PE router,
the IPv6 header is removed and the original IPv4 packet is forwarded
to the destination IPv4 network based on the outcome of the lookup in
the IPv4 routing table contained in the egress 4over6 PE router.
3.3. MP-BGP 4over6 Protocol Extensions
Each 4over6 PE router possesses an IPv4 interface connected to an
IPv4 access network(s). It can peer with other IPv4 routers using
IGP or BGP routing protocols to exchange local IPv4 routing
information. Routing information can also be installed on the 4over6
PE router using static configuration methods.
Each 4over6 PE also possesses at least one IPv6 interface to connect
it into the IPv6 transit backbone. The 4over6 PE typically uses IGP
routing protocols to exchange IPv6 backbone routing information with
other IPv6 P routers. The 4over6 PE router will also form an MP-iBGP
(Internal BGP) peering relationship with other 4over6 PE routers
connected to the IPv6 backbone network.
The use of MP-iBGP suggests that the participating 4over6 PE routers
that share a route reflector or form a full mesh of TCP connections
are contained in the same autonomous system (AS). This
implementation is in fact only deployed over a single AS. This was
not an intentional design constraint but rather reflected the single
AS topology of the CNGI-CERNET2 (China Next Generation Internet -
China Education and Research Network) national IPv6 backbone used in
the testing and deployment of this solution.
Wu, et al. Experimental [Page 7]
RFC 5747 4over6 March 2010
3.3.1. Receiving Routing Information from Local CE
When a 4over6 PE router learns routing information from the locally
attached IPv4 access networks, the 4over6 MP-iBGP entity should
process the information as follows:
1. Install and maintain local IPv4 routing information in the IPv4
routing database.
2. Install and maintain new entries in the encapsulation table.
Each entry should consist of the IPv4 prefix and the local IPv6
VIF address.
3. Advertise the new contents of the local encapsulation table in
the form of MP_REACH_NLRI update information to remote 4over6 PE
routers. The format of these updates is as follows:
* AFI = 1 (IPv4)
* SAFI = 67 (4over6)
* NLRI = IPv4 network prefix
* Network Address of Next Hop = IPv6 address of its 4over6 VIF
4. A new Subsequent Address Family Identifier (SAFI) BGP 4over6 (67)
has been assigned by IANA. We call a BGP update with a SAFI of
67 as 4over6 routing information.
3.3.2. Receiving 4over6 Routing Information from a Remote 4over6 PE
A local 4over6 PE router will receive MP_REACH_NLRI updates from
remote 4over6 routers and use that information to populate the local
encapsulation table and the BGP routing database. After validating
the correctness of the received attribute, the following procedures
are used to update the local encapsulation table and redistribute new
information to the local IPv4 routing table:
1. Validate the received BGP update packet as 4over6 routing
information by AFI = 1 (IPv4) and SAFI = 67 (4over6).
2. Extract the IPv4 network address from the NLRI field and install
as the IPv4 network prefix.
3. Extract the IPv6 address from the Network Address of the Next Hop
field and place that as an associated entry next to the IPv4
network index. (Note, this describes the update of the local
encapsulation table.)
Wu, et al. Experimental [Page 8]
RFC 5747 4over6 March 2010
4. Install and maintain a new entry in the encapsulation table with
the extracted IPv4 prefix and its corresponding IPv6 address.
5. Redistribute the new 4over6 routing information to the local IPv4
routing table. Set the destination network prefix as the
extracted IPv4 prefix, set the Next Hop as Null, and Set the
OUTPUT Interface as the 4over6 VIF on the local 4over6 PE router.
Therefore, when an ingress 4over6 PE router receives an IPv4 packet,
the lookup in its IPv4 routing table will have a result of the output
interface as the local 4over6 VIF, where the incoming IPv4 packet
will be encapsulated with a new IPv6 header, as indicated in the
encapsulation table.
4. 4over6 Deployment Experience
4.1. CNGI-CERNET2
A prototype of the 4over6 solution is implemented and deployed on
CNGI-CERNET2. CNGI-CERNET2 is one of the China Next Generation
Internet (CNGI) backbones, operated by the China Education and
Research Network (CERNET). CNGI-CERNET2 connects approximately 25
core nodes distributed in 20 cities in China at speeds of 2.5-10
Gb/s. The CNGI-CERNET2 backbone is IPv6-only with some attached
customer premise networks (CPNs) being dual stack. The CNGI-CERNET2
backbone, attached CNGI-CERNET2 CPNs, and CNGI-6IX Exchange all have
globally unique AS numbers. This IPv6 backbone is used to provide
transit IPv4 services for customer IPv4 networks connected via 4over6
PE routers to the backbone.
4.2. 4over6 Testbed on the CNGI-CERNET2 IPv6 Network
Figure 5 shows 4over6 deployment network topology.
Wu, et al. Experimental [Page 9]
RFC 5747 4over6 March 2010
+-----------------------------------------------------+
| IPv6 (CERNET2) |
| |
+-----------------------------------------------------+
| | | |
Tsinghua|Univ. Peking|Univ. SJTU| Southeast|Univ.
+------+ +------+ +------+ +------+
|4over6| ... |4over6| |4over6| ... |4over6|
|router| |router| |router| |router|
+------+ +------+ +------+ +------+
| | | |
| | | |
| | | |
+-----------+ +-----------+ +-----------+ +-----------+
|IPv4 access| ... |IPv4 access| |IPv4 access| ... |IPv4 access|
| network | | network | | network | | network |
+-----------+ +-----------+ +-----------+ +-----------+
|
+----------------------+
| IPv4 (Internet) |
| |
+----------------------+
Figure 5: 4over6 Deployment Network Topology
The IPv4-only access networks are equipped with servers and clients
running different applications. The 4over6 PE routers are deployed
at 8 x IPv6 nodes of CNGI-CERNET2, located in seven universities and
five cities across China. As suggested in Figure 5, some of the IPv4
access networks are connected to both IPv6 and IPv4 networks, and
others are only connected to the IPv6 backbone. In the deployment,
users in different IPv4 networks can communicate with each other
through 4over6 tunnels.
4.3. Deployment Experiences
A number of 4over6 PE routers were deployed and configured to support
the 4over6 transit solution. MP-BGP peerings were established, and
successful distribution of 4over6 SAFI information occurred.
Inspection of the BGP routing and encapsulation tables confirmed that
the correct entries were sent and received. ICMP ping traffic
indicated that IPv4 packets were successfully transiting the IPv6
backbone.
In addition, other application protocols were successfully tested per
the following:
Wu, et al. Experimental [Page 10]
RFC 5747 4over6 March 2010
o HTTP. A client running Internet Explorer in one IPv4 client
network was able to access and download multiple objects from an
HTTP server located in another IPv4 client network.
o P2P. BitComet software running on several PCs placed in different
IPv4 client networks were able to find each other and share files.
Other protocols, including FTP, SSH, IM (e.g., MSN, Google Talk), and
Multimedia Streaming, all functioned correctly.
5. Ongoing Experiment
Based on the above successful experiment, we are going to have
further experiments in the following two aspects.
1. Inter-AS 4over6
The above experiment is only deployed over a single AS. With the
growth of the network, there could be multiple ASes between the
edge networks. Specifically, the Next Hop field in MP-BGP
indicates the tunnel endpoint in the current 4over6 technology.
However, in the Inter-AS scenario, the tunnel endpoint needs to be
separated from the field of Next Hop. Moreover, since the
technology of 4over6 is deployed on the router running MP-BGP, the
supportability of 4over6 on each Autonomous System Border Router
(ASBR) will be a main concern in the Inter-AS experiment. We may
consider different situations: (1) Some ASBRs do not support
4over6; (2) ASBRs only support the 4over6 control plane (i.e., MP-
BGP extension of 4over6) rather than 4over6 data plane; (3) ASBRs
support both the control plane and the data plane for 4over6.
2. Multicast 4over6
The current 4over6 technology only supports unicast routing and
data forwarding. With the deployment of network-layer multicast
in multiple IPv4 edge networks, we need to extend the 4over6
technology to support multicast including both multicast tree
manipulation on the control plane and multicast traffic forwarding
on the data plane. Based on the current unicast 4over6 technology
providing the unicast connectivity of edge networks over the
backbone in another address family, the multicast 4over6 will
focus on the mapping technologies between the multicast groups in
the different address families.
Wu, et al. Experimental [Page 11]
RFC 5747 4over6 March 2010
6. Relationship to Softwire Mesh Effort
The 4over6 solution was presented at the IETF Softwires Working Group
Interim meeting in Hong Kong in January 2006. The existence of this
large-scale implementation and deployment clearly showed that MP-BGP
could be employed to support tunnel setup in a scalable fashion
across an IPv6 backbone. Perhaps most important was the use-case
presented -- an IPv6 backbone that offers transit to attached client
IPv4 networks.
The 4over6 solution can be viewed as a precursor to the Softwire Mesh
Framework proposed in the softwire problem statement [RFC4925].
However, there are several differences with this solution and the
effort that emerged from the Softwires Working Group called "softwire
Mesh Framework" [RFC5565] and the related solutions [RFC5512]
[RFC5549].
o MP-BGP Extensions. 4over6 employs a new SAFI (BGP 4over6) to
convey client IPv4 prefixes between 4over6 PE routers. Softwire
Mesh retains the original AFI-SAFI designations, but it uses a
modified MP_REACH_NLRI format to convey IPv4 Network Layer
Reachability Information (NLRI) prefix information with an IPv6
next_hop address [RFC5549].
o Encapsulation. 4over6 assumes IP-in-IP or it is possible to
configure Generic Routing Encapsulation (GRE). Softwires uses
those two scenarios configured locally or for IP headers that
require dynamic updating. As a result, the BGP encapsulation SAFI
is introduced in [RFC5512].
o Multicast. The basic 4over6 solution only implemented unicast
communications. The multicast communications are specified in the
Softwire Mesh Framework and are also supported by the multicast
extension of 4over6.
o Use-Cases. The 4over6 solution in this document specifies the
4over6 use-case, which is also pretty easy to extend for the use-
case of 6over4. The Softwire Mesh Framework supports both 4over6
and 6over4.
7. IANA Considerations
A new SAFI value (67) has been assigned by IANA for the BGP 4over6
SAFI.
Wu, et al. Experimental [Page 12]
RFC 5747 4over6 March 2010
8. Security Considerations
Tunneling mechanisms, especially automatic ones, often have potential
problems of Distributed Denial of Service (DDoS) attacks on the
tunnel entry-point or tunnel exit-point. As the advantage, the BGP
4over6 extension doesn't allocate resources to a single flow or
maintain the state for a flow. However, since the IPv6 tunnel
endpoints are globally reachable IPv6 addresses, it would be trivial
to spoof IPv4 packets by encapsulating and sending them over IPv6 to
the tunnel interface. This could bypass IPv4 Reverse Path Forwarding
(RPF) or other antispoofing techniques. Also, any IPv4 filters may
be bypassed.
An iBGP peering relationship may be maintained over IPsec or other
secure communications.
9. Conclusion
The emerging and growing deployment of IPv6 networks, in particular,
IPv6 backbone networks, will introduce cases where connectivity with
IPv4 networks is desired. Some IPv6 backbones will need to offer
transit services to attached IPv4 access networks. The 4over6
solution outlined in this document supports such a capability through
an extension to MP-BGP to convey IPv4 routing information along with
an associated IPv6 address. Basic IP encapsulation is used in the
data plane as IPv4 packets are tunneled through the IPv6 backbone.
An actual implementation has been developed and deployed on the CNGI-
CERNET2 IPv6 backbone.
10. Acknowledgements
During the design procedure of the 4over6 framework and definition of
BGP-MP 4over6 extension, Professor Ke Xu gave the authors many
valuable comments. The support of the IETF Softwires WG is also
gratefully acknowledged with special thanks to David Ward, Alain
Durand, and Mark Townsley for their rich experience and knowledge in
this field. Yakov Rekhter provided helpful comments and advice.
Mark Townsley reviewed this document carefully and gave the authors a
lot of valuable comments, which were very important for improving
this document.
The deployment and test for the prototype system was conducted among
seven universities -- namely, Tsinghua University, Peking University,
Beijing University of Post and Telecommunications, Shanghai Jiaotong
University, Huazhong University of Science and Technology, Southeast
Wu, et al. Experimental [Page 13]
RFC 5747 4over6 March 2010
University, and South China University of Technology. The authors
would like to thank everyone involved in this effort at these
universities.
11. Normative References
[RFC2473] Conta, A. and S. Deering, "Generic Packet Tunneling in
IPv6 Specification", RFC 2473, December 1998.
[RFC4271] Rekhter, Y., Li, T., and S. Hares, "A Border Gateway
Protocol 4 (BGP-4)", RFC 4271, January 2006.
[RFC4760] Bates, T., Chandra, R., Katz, D., and Y. Rekhter,
"Multiprotocol Extensions for BGP-4", RFC 4760,
January 2007.
[RFC4925] Li, X., Dawkins, S., Ward, D., and A. Durand, "Softwire
Problem Statement", RFC 4925, July 2007.
[RFC5512] Mohapatra, P. and E. Rosen, "The BGP Encapsulation
Subsequent Address Family Identifier (SAFI) and the BGP
Tunnel Encapsulation Attribute", RFC 5512, April 2009.
[RFC5549] Le Faucheur, F. and E. Rosen, "Advertising IPv4 Network
Layer Reachability Information with an IPv6 Next Hop",
RFC 5549, May 2009.
[RFC5565] Wu, J., Cui, Y., Metz, C., and E. Rosen, "Softwire Mesh
Framework", RFC 5565, June 2009.
Wu, et al. Experimental [Page 14]
RFC 5747 4over6 March 2010
Authors' Addresses
Jianping Wu
Tsinghua University
Department of Computer Science, Tsinghua University
Beijing 100084
P.R. China
Phone: +86-10-6278-5983
EMail:
[email protected]
Yong Cui
Tsinghua University
Department of Computer Science, Tsinghua University
Beijing 100084
P.R. China
Phone: +86-10-6278-5822
EMail:
[email protected]
Xing Li
Tsinghua University
Department of Electronic Engineering, Tsinghua University
Beijing 100084
P.R. China
Phone: +86-10-6278-5983
EMail:
[email protected]
Mingwei Xu
Tsinghua University
Department of Computer Science, Tsinghua University
Beijing 100084
P.R. China
Phone: +86-10-6278-5822
EMail:
[email protected]
Chris Metz
Cisco Systems, Inc.
3700 Cisco Way
San Jose, CA 95134
USA
EMail:
[email protected]
Wu, et al. Experimental [Page 15]
