Soft Partitioning Flow Tables for Virtual Networking in Multi-tenant Software-Defined Networks

• Main Authors: Chen, Jian-Hao, 陳建豪
• Other Authors: Lin, Ying-Dar
• Format: Others
• Language: en_US
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/2pfv38

碩士 === 國立交通大學 === 資訊科學與工程研究所 === 104 === In a multi-tenancy SDN environment, physical devices such as switches are shared among tenants. Proxy architecture, where a proxy controller sits between tenant controllers and SDN switches, is often adopted to manage resources such as flow table in each switch among different tenants. Hard partitioning of a flow table allocates dedicated flow entries to each tenant, but allocated flow entries are wasted if the tenant does not actually use them. On the other hand, soft partitioning strategy shares available flow entries among tenants, resulting in high utilization and a resource monopoly problem, i.e., flow entries dominated by some heavy-traffic tenants. To achieve high flow table utilization and avoid the resource monopoly problem, we propose Soft-Partitioning Resource Manager (SPRM) to manage the flow table resources in multi-tenancy SDN environment. SPRM has three main concepts: (1) hybrid mode: SPRM operates in two modes: in-line mode where the proxy controller submits flow modifications to SDN switches after it performs necessary checks, and sniff mode where these two operations are proceeded in parallel. SPRM operates in sniff mode when there are free flow entries to efficiently reduce flow modification latency; (2) LRU-PTR (least recently used with partial timeout reset) replacement: SPRM, using partial timeout before real idle timeout, lets the switches report the age of flow entries, so flow entries can be replaced by LRU-PTR replacement if needed; and (3) usage range: SPRM sets lower and upper bounds for each tenant, and maintains the tenant flow entry usage locating between these bounds. Experimental results show that 100% rejections, i.e., flow modification failures, and 99% packet_in events are reduced while flow modification latency is decreased by 30%, compared to hard partitioning.