A Survey on TCAM Storage Optimization for SDN Switch Forwarding Rules

Software Defined Networking (SDN) makes network control more convenient and more flexible by decoupling the control plane and the data plane. The controller interacts with switches through specific control protocols such as OpenFlow, and rules for forwarding packets will be installed or updated in the switches. Ternary Content Addressable Memory (TCAM) is the most widely used storage medium for forwarding rules in modern SDN switches due to its fast lookup. TCAM supports parallel lookup and outputs the results in one clock cycle. Besides, TCAM can store rules with wildcard directly, which is common in OpenFlow networks. However, TCAM suffers the shortcomings of high cost, high power consumption, and the problem of range expansion when storing range-field-contained forwarding rules. As is known to all, OpenFlow has been the most widely used control protocol at present, and the OpenFlow protocol standard has specified more than 40 match fields according to its latest version. The number of match fields is still growing to achieve high quality network performance. Therefore, forwarding rules in an OpenFlow switch specify longer bits and occupy more room in TCAM than those in a traditional L2 or L3 switch. Therefore, the number of forwarding rules that TCAM can store is very limited, especially the OpenFlow forwarding rules with various match fields. It becomes a bottleneck of software-defined network development. In order to efficiently utilize the limited TCAM storage resources for forwarding rules, it is necessary to optimize the storage problem of forwarding rules in SDN. This paper mainly analyses and summarizes TCAM storage optimization mechanisms of forwarding rules in SDN from such four perspectives as forwarding rule storage architecture optimization, compression of forwarding rules locally and globally, management of forwarding rules with the participation of controller(s). First, forwarding rule storage architecture optimization aims at improving lookup circuit structure in switches, replacing TCAM with better storage media, or applying a mix storage scheme of multiple storage medium. Most methods involve hardware modification. Storing multiple sub-tables rather than the original rule table can improve the storage utilization in TCAM. It is often necessary to maintain information of jump instruction. Second, we can adopt compression algorithms to reduce the number of rules under the premise of keeping rule table semantics unchanged. Compression algorithms need to be effective and efficient, and can compress rule table with high dimension. Third, rule table storage pressure can be relieved by optimizing the routing of flows globally in SDN networks. For example, one rule can be multiplexed by multiple flows in specific network segments, through which the number of rules is reduced. Algorithms are developed in the direction of reducing the number of rules among the entire network. Fourth, Management of forwarding rules with the participation of controller (s) includes caching of forwarding rules or flow table overflow control through specifically designed modules. The module is designed in controllers or additional layer(s) between the control and data plane. However, none of the above four solutions can solve the TCAM storage problem thoroughly. Each of them suffers certain limitations such as overhead of hardware modification, high algorithm complexity, algorithm ineffectiveness and so on. Therefore, we discussed the comprehensive schemes of forwarding rule optimization suitable for future SDN at last. © 2021, Science Press. All right reserved.