Information Technology Reference
In-Depth Information
which is implemented by Bloomfilter[17]. When data packets arrive, services
modify labels, choose the output port, and delete the chunk number from the
member set of labels and ports. The Label table and the Cast table implement
the forwarding of interest packets and the generation of data packets together,
which correspond the functions of PIT and CS in NDN, as illustrated in Fig.4.
5 Prototype of LabelCast
5.1 Implementation of the Prototype
Based on the general-purpose multi-core processor FT1000 and Network Pro-
cessing Engine, NPE, we implement a prototype system to support LabelCast.
The Label table is implemented in NPE to schedule packet forwarding operations
and packet output control instructions in data plane. While taking the advan-
tage of the computing and storage resources within FT1000 connected through
the system bus PCI-E, we design the Cast table to implement a protocol and
status related packet processing services.
In the prototype system, the eXtended Linux container, XLC, is designed to
implement the virtualization and allocation of computing and storage resources,
which simplify the development of user-defined services. Based on a scalable
resource container, researchers could design and implement the custom packet
forwarding and processing services, and in the development of services the appli-
cation could call the system components and libraries provided by the Labelcast
prototype system to accelerate applications development.
5.2 PacketDirect IO Performance
Labelcast data plane provides the basic operations of packet forwarding and
services of packets processing. Basic forwarding operations in data plane are im-
plemented in NPE, to provide the light semantics actions of general packet pro-
cessing, including look-up table, modifying the basic packet options and output
control. Packets processing are implemented by the services running on FT1000
being tightly coupled with the NPE, to implement network protocol semantics
or the state-related deep packet processing services in data path, such as the
replacement of specific fields of packets, packets caching and the calculation of
route based on the matrix of the network layer reaching information. Therefore,
in the prototype system the system bus between NPE and FT1000 becomes the
key to improve the system performance.
In the design and implementation of the prototype system, an ecient Packet-
Direct mechanism was designed and implemented to improve system throughput
of the transport mechanism between NPE and FT1000. In experiments network
tester transmitting packets to NPE via two 10Gbps ports, testing the Pack-
etDirect IO performance of the prototype in the case of different number of
threads and packets size, the results are shown in Fig.5. With the number of
forwarding threads increasing, the performance of processing has a certain up-
grade under different packets size; 1 thread implements line-rate forwarding of
