Information Technology Reference
In-Depth Information
NOS peers use TCP to setup connections. After a TCP connection is established,
the first message sent is the OPEN message. Some parameters such as timers should be
negotiated in the OPEN message. After the OPEN message is confirmed, NOS will
keeps sending out KEEPALIVE message. EWBridge defines a timer for the
KEEPALIVE message. If one peer has not been receiving KEEPALIVE packet from a
certain NOS, this peer will think the NOS has failed, and will release this connection.
All the errors will lead to idle state and NOTIFICAION will be sent to all the peers.
NOS also can use the VIEW-REFRESH message to fetch network view information
from its certain peers.
In the normal condition each NOS keeps sending out KEEPALIVE messages to its
peers in parallel. So it can keep in connection with all its peers. Once network view
updates within its domain appear, it can directly send UPDATE file out to its peers in
parallel without re-setup TCP connections. Furthermore, each UPDATE file can carry
multiple UPDATE messages. In such way, EWBridge can achieve high speed data
exchange.
4
Implementation and Evaluation
4.1
Implementation and Experiment Results
EWBridge has been implemented in two open-source NOSes: Floodlight and NOX.
We use Mininet [3] to emulate two data centers. One is located in Beijing, China, and
the other is located in California, USA, as shown in Figure 5. The OpenFlow switches
in network are Open vSwitches [11].We divide the entire network of Beijing data cen-
ter into four sub-networks and divide the California data center into three sub-networks.
Each sub-network runs one NOS and one database. We run a centralized register center
server for EWBridge in sub-network 6 in California. We also run a topology viewer
application and a video server process on this server.
Fig. 5.
Experiment Topology
In the first experiment, all the NOSes are Floodlight software and all databases are
Cassandra [12] DHT databases. We write and run a topology viewer application to fetch
the entire topology of all the sub-networks. Then in the second experiment, the control-
lers in sub-network 1, 3, 5, 7 run Floodlight and the controllers in sub-network 2, 4, 6
