MS research page cited 54
Increased use of demanding network applications, as well as the increase of unwanted network traffic in the form of DDoS attacks, are putting new pressures on service providers to meet the expectations of customers in terms of network availability and performance.
dynamic connectivity management, which we broadly define as the ability to dynamically manage how and where traffic flows across a network.
simplification introduced by a centralized Intelligent Route Service Control Point (IRSCP)
… increased occurrence of distributed denial of service (DDoS) attacks likewise require more sophisticated and responsive network management practices from providers. We broadly define this timely control of how traffic flows through a network as dynamic connectivity management.
In the early 1980’s the circuit switched voice network underwent a revolution with the introduction of a technology called the Network Control Point (NCP) [1].
our work … enable (arbitrary) external information to inform the route selection process
The IRSCP is a logically centralized network control element, i.e., it
takes part in “control plane” functions but is
Figure 1 shows two forms of input into the IRSCP. First is direct operator input, for example when a task like blackholing of DDoS traffic is performed. The second IRSCP input is what we broadly call “network intelligence” and represents the fact that the IRSCP platform allows external information to directly impact the routing process.
critically important advantage because DDoS attacks are in fact not that distributed… a recent study [9] showed that for DDoS attacks observed in an ISP network, over a four week period, only 0.1% of ingress interfaces contributed more than 90% of the DDoS traffic volume.
ISPs routinely perform planned maintenance on routers to replace faulty hardware or install new router software.
when customer-edge (CE) routers are dual homed to two provider-edge (PE) routers
for all prefixes advertised by the dryout-router, if those prefixes are available from another router, make them more preferred.
common practice to- day to realize dryout is to change the IGP weight of selected links in the network to force traffic off the dryout router
The provider network has no knowledge of these customer goals and simply route traffic across the backbone network according to default shortest path behavior.
a significant portion of the traffic destined to the data center (or customer network) is entering the IRSCP-enabled network from AS 1. Assuming that all IGP links weights are the same, both PE3 and PE4 will prefer to reach the data center using the routes advertised by PE2. Either way the net result is that the link between PE2 and CE will carry most of the traffic while the link between PE1 and CE will be mostly idle.
common problem for providers and customers alike … a CDF of the traffic ratio between the most loaded link and the least loaded link for each multi-homed customer in a large ISP over a typical day.