Reseach On Micro-loop Avoidance Mathod And Congestion Free Methon During Network Update

Network operator had to update the existing network to copy with the growing traffic and increasing requirement for quality of services. However, a series of problems which lead to network performance degradation often happened during network information re-configuration. This paper studied typical issues of IP network update and SDN network update and proposed corresponding solutions for them.IP network update would trigger routing re-convergence. However, as all routers could not receive the new link state and enable the post-convergence forwarding table simultaneously, it might give packets chances to fall into a temporary loop formed by forwarding paths, namely micro-loop. This micro-loop problem had been the main issue due to network potential risk of increased latency and packet loss it brought.Software Defined Network(SDN) update was generally realized by re-configuring flow tables. However, it is difficult even impossible to force all SDN switches to perform update action in a strict synchronized way. Thus SDN was likely to see inconsistency for packets forwarded along mixed rules combining the old rule with the new one. The inconsistency problem might further trigger black holes, micro-loop and security issues. Because of the powerful flow-control ability of SDN and the high quality of service requirement of SDN applications, how to ensure consistency had been widely concerned. However, even consistency was satisfied, SDN still could not promise all flows followed the old flow table or the new one. So some hot links would probably suffered from transient congestion and traffic loss due to ’new-rule’ traffic and ’old-rule’ traffic were racing for bandwidth.The work done by this paper included the following aspects.1) This paper studied and analyzed two mainstream IP micro-loop avoidance technologies: OFIB and OMA. OFIB updated the network by directly controlling the FIB update time of each router. OMA progressively adjusted the link weight to the target in multiple stages to realize network updates. And each phase adjusted link metric to an intermediate one which ensured loop free of routing convergence. Compared with OFIB, OMA had gained boarder attentions and recognitions due to its better expansibility and easy implementation.2) For the high computational complexity of existing micro-loop avoidance solutions based on OMA, this paper proposed a rather low-complexity micro-loop avoidance algorithm(LCMLAA). LCMLAA defined "metric range" and computed minimum number of middle metrics by taking advantage of range overlapping directly.3) This paper studied and analyzed three mainstream SDN update mechanisms: update mechanism based on sequence number, update mechanism based on store-and-forward and update mechanism based on the version number. The update mechanism based on sequence number designed a sequence information table for all flow entries and completed update by implementing the instruction sets in that table. The update mechanism based on store-and-forward forwarded some packets to controller cache during update and then issued those packets back to data plane after update. The update mechanism based on version made use of additional label to distinguish new rules and old ones. Compared with the other two update mechanisms, update mechanism based on version number was the only one which could ensure network consistency.4) For update mechanism based on the version number failed to solve link congestion problem, this paper proposed a congestion free update strategy which scheduled a reasonable order to guide all traffic flows transfer from old rule to new one so that any link was freedom from being congested. Based on the above strategy, this paper proposed congestion avoidance fast update algorithm(CAFUA) and congestion avoidance fast update heuristic algorithm(ATOMIP). The former studied node update order problem which took node as smallest update unit(i.e., all flow table entries on the same node were confined to be updated in one round), and targeted to reduce the control signaling overhead between the central controller and switches as well as the impact of switches’ capability differences on network performance. The latter studied flow update order problem which took single traffic flow as smallest update unit(i.e., incoming flows from the same ingress node were allowed to be updated in multiple rounds), and targeted to achieve fewer update rounds and improve adaptability of large-scale network update by taking advantage of the fine control ability of SDN.